feat(tools): Enhance guest tools installation and management

- Added new API endpoints for tool catalog and installation management.
- Updated usage documentation to include new tool installation features.
- Implemented asynchronous tool installation with job tracking.
- Introduced integration tests for tool installation in disposable VMs.
- Enhanced dashboard UI to reflect tool installation status and usage instructions.
- Improved error handling and logging for tool installation processes.
- Added support for multiple Python versions and Conda environments.
- Refactored tool management logic to ensure better state handling and job persistence.

DEVELOPER.md

@@ -19,7 +19,7 @@ expose health, installation, VM lifecycle controls, and managed port-policy
 controls, host diagnostics, bounded VM-log tails, and fixed-target bulk
 operations, and bounded guest command execution over vsock. Interactive SSH remains CLI-only.
 
-There is no autostart or background reconciler, multi-host control plane,
+There is no general VM autostart or background reconciler, multi-host control plane,
 tenancy/quotas, or UVM-provided Jailer/cgroup/namespace hardening. Private port
 filtering is not public-port publishing or a hardened multi-tenant boundary;
 see Current Production Gaps for the remaining scope limits.
@@ -95,6 +95,10 @@ Packet integration tests must remain inside disposable isolated namespaces.
 │   ├── guest_agent.py        # Standalone guest daemon, job scopes, and bounded wire protocol
 │   ├── guest_exec.py         # Host-side guest sessions and execution results
 │   ├── guest_terminal.py     # Async PTY transport over a process-pinned guest connection
+│   ├── tool_catalog.py       # Allowlisted tools, Python selections, and pinned release recipes
+│   ├── guest_tools.py        # Standalone guest-only package/download installer and probes
+│   ├── tool_store.py         # Atomic per-VM job history and inventory observations
+│   ├── tools.py              # Bounded background tool installation workers
 │   ├── network.py            # IP/MAC/TAP/bridge/NAT/forward rules
 │   ├── firewall.py           # Owned nft bridge tables and offline conntrack cleanup
 │   ├── lifecycle.py          # Lifecycle, bulk operations, recovery, port intent
@@ -106,6 +110,7 @@ Packet integration tests must remain inside disposable isolated namespaces.
 │   │   ├── exit.py           # Authenticated server shutdown
 │   │   ├── health.py         # Liveness endpoint
 │   │   ├── installation.py   # Installation endpoint
+│   │   ├── tools.py          # Catalog, inventory, and installation job endpoints
 │   │   └── vms.py            # VM lifecycle endpoints
 │   └── firecracker/
 │       ├── config.py         # Firecracker JSON and boot arguments
@@ -218,6 +223,8 @@ process:
 | FirecrackerProcessManager | `FirecrackerProcessManager` | VMM process control |
 | LifecycleService | `LifecycleService` | Ordered VM resource changes and rollback |
 | Installer | `Installer` | Host packages and artifacts |
+| ToolService | `ToolService` | Browser-independent guest installation queue and workers |
+| ToolJobStore | Constructed by lifecycle/tool services | Per-VM installation intent, results, and cached inventory |
 
 Composition can disable command tracing with `emit=None`; `create_api()` does
 so when it constructs its own application. The CLI supplies an emitter that
@@ -245,10 +252,12 @@ Default runtime storage is `/var/lib/uvm`:
 │       ├── config.json
 │       ├── firecracker.log
 │       ├── firecracker.sock
+│       ├── tools.json
 │       └── vsock.sock
 ├── state.json
 ├── state.lock
 ├── operations.lock
+├── tool-worker.lock
 └── integrity.json
 ```
 
@@ -272,6 +281,8 @@ state file.
 
 `VmSpec` is the validated create request. `VmRecord` is the persisted local
 record. Do not make API models or CLI namespaces the persistence format.
+`VmSpec.tools` is a tuple of validated `ToolSelection` values. Tool job state
+uses its own schema-1 `vms/<id>/tools.json`; the VM registry remains schema 3.
 
 | Record Fields | Purpose |
 |---|---|
@@ -363,7 +374,8 @@ a ruleset/attachment snapshot, not application health or a reachability test.
 3. Reserve ID/name/IP/MAC/ports with `status=starting`, `operation=preparing`.
 4. Create the owned mode-0700 runtime directory, sparse-copy a mode-0600 private
    disk, resize ext4 to the requested storage capacity, provision credentials/host
-   keys and the managed guest network profile, then persist `disk_ready=True`.
+   keys, the managed guest network profile, and the agent, then persist `disk_ready=True`.
+   If tools were selected, persist their queued job in the owned runtime directory.
 5. Ensure bridge/egress networking and prepare an owned, attached TAP **down**.
 6. Apply/read back the nftables bridge policy, then activate the TAP.
 7. Write Firecracker config; persist `operation=spawning`, `spawn_pending=True`,
@@ -371,10 +383,13 @@ a ruleset/attachment snapshot, not application health or a reachability test.
 8. Publish PID/start time/boot ID through `on_spawn` before socket readiness
    or API configuration; clear the pending-spawn guard.
 9. Configure Firecracker and `InstanceStart`, verify policy enforcement again,
-   then save `running` with no pending operation.
+then save `running` with no pending operation.
 
 This records both intent and acquired ownership across crash windows. Never
 make a managed TAP usable before enforcement or mark a partial disk ready.
+Tool workers claim the queued intent after this operation releases its lock.
+Tool installation errors are job results on the successfully created VM and
+never enter create's resource rollback path.
 
 ### Rollback Rules
 
@@ -505,7 +520,8 @@ snapshot produced before CLI confirmation. It reloads by **exact ID**, not a
 reusable name/IP. Results are `succeeded`, `failed`, `skipped` (already absent),
 or `interrupted`; ordinary failures do not undo other successes or stop later
 targets. Interruptions skip the remaining targets. Read-only views and API
-startup never reconcile; there is no autostart or background repair loop.
+startup never reconcile lifecycle or firewall intent. The tool worker resumes
+explicitly queued installation requests, as described under Guest Tools below.
 
 ## 9. Firecracker Integration
 
@@ -792,7 +808,7 @@ do not import its unimplemented topology checks into current documentation.
 
 ## 12. Installer and Host Commands
 
-`Installer` is the only owner of apt installation, Firecracker artifact
+`Installer` is the owner of host apt installation, Firecracker artifact
 handling, downloads, and integrity-manifest publication.
 
 `Settings` detects and normalizes the host architecture at construction time
@@ -893,7 +909,7 @@ Nest. Both server entry points expose the UVM management routes below.
 | `GET` | `/diagnostics` | `diagnostics.py` | Host readiness checks |
 | `POST` | `/install` | `installation.py` | `{ "force": false }` body |
 | `GET` | `/vms` | `vms.py` | List persisted VM records and observed status |
-| `POST` | `/vms` | `vms.py` | `cpu`, `ram`, `storage`, `guest_ip`, `username`, `password`, `name`, `ports`, `no_default_ports` |
+| `POST` | `/vms` | `vms.py` | `cpu`, `ram`, `storage`, `guest_ip`, `username`, `password`, `name`, `ports`, `no_default_ports`, `tools` |
 | `POST` | `/vms/actions` | `vms.py` | Fixed-ID snapshot bulk lifecycle action |
 | `GET` | `/vms/{identifier}` | `vms.py` | Inspect by exact ID, name, or guest IP |
 | `POST` | `/vms/{identifier}/start` | `vms.py` | Boot a retained private disk |
@@ -905,10 +921,16 @@ Nest. Both server entry points expose the UVM management routes below.
 | `POST` | `/vms/{vm_id}/ports/sync` | `vms.py` | Exact ID only; reapply managed policy |
 | `POST` | `/vms/{vm_id}/exec` | `vms.py` | Exact ID; `command`, strict `sudo`, bounded `timeout_seconds` |
 | `POST` | `/vms/{vm_id}/agent/install` | `vms.py` | Exact ID; install/upgrade the agent on a stopped disk |
+| `GET` | `/tools` | `tools.py` | Authenticated catalog and supported Python versions |
+| `GET` | `/vms/{vm_id}/tools` | `tools.py` | Cached inventory/history; `refresh=true` probes the running guest |
+| `POST` | `/vms/{vm_id}/tools/install` | `tools.py` | Exact ID; `tools`, `start_if_stopped`, `restart`, optional `request_id`; HTTP 202 |
+| `GET` | `/vms/{vm_id}/tools/jobs/{job_id}` | `tools.py` | Exact VM-scoped installation job and bounded output |
 | `GET` | `/vms/{identifier}/logs` | `vms.py` | Bounded non-following log tail |
 | `GET` | `/vms/{vm_id}/sessions` | `nest/session_routes.py` | Saved session list (Nest only) |
 | `POST` | `/vms/{vm_id}/sessions` | `nest/session_routes.py` | Launch a guest PTY; optional rows/cols in a JSON object (Nest only) |
 | `GET` | `/vms/{vm_id}/sessions/{session_id}` | `nest/session_routes.py` | VM-scoped session metadata (Nest only) |
+| `PATCH` | `/vms/{vm_id}/sessions/{session_id}` | `nest/session_routes.py` | Rename via a strict `name` body; identity retained (Nest only) |
+| `DELETE` | `/vms/{vm_id}/sessions/{session_id}` | `nest/session_routes.py` | Close channel, join output reader, delete history; 204 (Nest only) |
 | `POST` | `/vms/{vm_id}/sessions/{session_id}/close` | `nest/session_routes.py` | End a shell; retain history (Nest only) |
 | `WS` | `/vms/{vm_id}/sessions/{session_id}/terminal` | `nest/session_routes.py` | Authenticated replay/input/resize (Nest only) |
 
@@ -1028,6 +1050,78 @@ device and uses CID 3. The process adapter retains both sockets while identity
 is uncertain and removes them only after confirmed VMM exit. IPv4 port policies
 do not govern this host-only management channel.
 
+### Guest Tools
+
+`tool_catalog.py` is the source of truth for catalog metadata, normalized
+`ToolSelection` values, and installer recipes. The API accepts catalog IDs and
+explicit Python minors; executable text, download URLs, paths, and other runtime
+versions are not request fields. There are ten tools and twelve selectable
+entries when Python 3.11/3.12/3.13 are expanded. Update publisher versions, URLs,
+and SHA-256 values together when changing a downloaded default.
+
+`ToolService` owns two bounded worker threads by default. The shared UVM server
+lifespan starts/stops them, so both Nest and the standalone API process the same
+queue. An exclusive, mode-0600 `$UVM_BASE/tool-worker.lock` permits one management
+server to own these workers per registry. Construction alone starts no threads.
+Queued requests are dispatched on server startup; prior in-flight jobs are
+marked interrupted with unknown outcomes after ownership has been acquired.
+
+`LifecycleService.queue_tools()` validates an exact VM ID and persists a request
+under the operation lock. Only one queued/active batch is allowed per VM.
+Identical active requests are deduplicated. Optional request IDs also identify
+completed jobs until they age out of the ten-job history; conflicting reuse is
+an error. `prepare_tool_install()` performs an explicitly requested start or
+restart under the normal lifecycle lock and returns the selected process identity.
+Creation writes its tool job after disk provisioning and before boot.
+
+Workers wait up to `Settings.tool_ready_timeout_s` (120 seconds) for the agent.
+Each tool has `Settings.tool_install_timeout_s` (1,200 seconds), bounded by the
+existing guest-exec protocol maximum. Readiness connection failures can be
+retried before submission; an uncertain installation result is never replayed.
+Each step passes `expected_identity` to `LifecycleService.execute()`, so a restart
+between steps cannot retarget the batch to a new VMM. The guest boot ID also
+detects in-guest reboots. Guest downloads and package waits hold no host operation
+lock. Shutdown closes tracked waiting sockets, preserves interruption results,
+and keeps worker ownership until outstanding host work has finished.
+
+`guest_tools.py` is a standalone stdlib-only Python 3.10+ payload. Its source and
+server-generated JSON recipes are shell-quoted into the bounded guest exec
+request. Importing it on the host performs no installation. In the guest it:
+
+- Checks Ubuntu 22.04 and the expected x86_64/ARM64 architecture before mutation.
+- Serializes installation through `/run/uvm-tools.lock`, including inherited
+  lock ownership during package commands. A previous disconnected job can finish
+  in the guest, and this lock prevents overlapping retries.
+- Uses non-interactive apt with bounded dpkg-lock waits and authenticated sources.
+  A pending marker permits a retry to finish its interrupted dpkg configuration.
+  PHP's repository key is checked against a pinned fingerprint and scoped with
+  `signed-by`.
+- Checks pinned hashes before executing/extracting downloads. Archive extraction
+  validates paths/links, file types, and size bounds. Versioned payloads live in
+  `/opt/uvm-tools`; executable links and login-shell profiles have an ownership
+  manifest that detects guest edits.
+- Checks command versions and companions as the configured guest account, including
+  npm/npx, Python pip/venv imports, and both Java tools. Python has versioned
+  executables; Conda uses per-user environments/caches. Get-started commands use
+  explicit POSIX shell hooks for activation in Nest's non-login shells.
+- Captures a bounded 16 KiB output tail and returns structured JSON when each
+  step finishes. An ordinary recipe failure allows later tools to be attempted.
+
+`ToolJobStore` keeps cached inventory and the last ten jobs in a mode-0600,
+schema-1 `tools.json` in the existing private VM runtime directory. Atomic
+replacement/fsync and short operation-lock transactions prevent races with
+deletion or stale lifecycle record saves. Store writes require a still-registered,
+owned VM directory and never recreate a deleted VM. VM deletion removes this
+history with the runtime. Full inventory refreshes update `checked_at`; step
+results also update individual tool observations. GET without `refresh=true`
+returns cached observations, including the `stale` flag.
+
+The dashboard's `tools.js` owns create selections, the VM Tools tab, job polling,
+history, retries, and output rendering. Active jobs poll every 1.5 seconds while
+visible; idle views poll cached status every five seconds. New selections use the
+shared catalog. Navigation/logout invalidate in-flight UI responses, and tool
+output is escaped before rendering.
+
 ### Nest Terminal Sessions
 
 Nest owns session HTTP/WebSocket routes and history. UVM owns the process-pinned
@@ -1051,8 +1145,12 @@ window-size changes are handled by the guest terminal driver. Normal exit,
 explicit close, or host connection loss cleans up the job cgroup and processes.
 
 `SessionManager` owns background output readers independently of WebSocket
-attachments. Creating a shell is shielded from an HTTP disconnect; its connection
-is either tracked or closed. Browser navigation, reload, and logout detach only
+attachments. Accepted creation, rename, close, and delete operations are tracked
+and shielded from HTTP disconnects. Creation and termination share a session
+lock, so deleting a starting session waits for its channel to be tracked and
+then closes it. Deletion joins the output reader and its pending writes before
+atomically removing the metadata and transcript rows. Browser navigation,
+reload, and logout detach only
 the browser. Nest shutdown closes guest connections and marks live sessions
 interrupted. In-flight transcript writes finish before releasing store ownership.
 
@@ -1062,7 +1160,11 @@ directory, mode-0600 files, an exclusive owner lock, and SQLite transactions
 protect history. On reopening after a server restart, unfinished records become
 interrupted. Session lookups always include the exact VM ID. Output is capped at
 the newest 8 MiB per session with byte offsets and an explicit truncation marker;
-session metadata is retained. Only PTY output is recorded, not a separate input
+session metadata is retained until explicit deletion. Session names are trimmed,
+1–80 printable characters, and are display labels rather than identifiers.
+Renaming updates only the name and modification time; generated default names
+avoid labels still present after renames or deletions. Only PTY output is
+recorded, not a separate input
 or password log. Nest permits 32 active sessions; guest terminals share the
 agent's eight connection slots with one-shot jobs.
 
@@ -1074,6 +1176,9 @@ offset after xterm renders the chunk. Only one chunk is outstanding, providing
 backpressure. Reattachment resumes from an acknowledged offset; a truncated
 prefix resets the viewer. Closed sessions use the same channel for read-only
 replay. WebSocket transport support comes from the `websockets` dependency.
+Renames notify attached viewers through updated session metadata. A deleted
+session sends a `deleted` event to authenticated attachments; polling also clears
+stale selections for completed sessions whose replay connection already closed.
 
 The dashboard's Sessions tab and nested hash routes belong to the selected VM.
 `sessions.js` owns attachment teardown, replay, reconnects, and the terminal
@@ -1182,7 +1287,18 @@ NODE_PATH=/tmp/nest-browser/node_modules \
 ```
 
 This checks launch/input, reload/reconnect, session switching, per-VM isolation,
-closed-session replay, and the mobile layout. It does not boot a real microVM.
+closed-session replay, rename persistence, active/ended deletion, cancellation,
+multi-tab updates, and the mobile layout. It does not boot a real microVM.
+
+With the same browser dependencies, `node tests/browser_tools.cjs` checks creation
+selections, persisted background jobs across reload, installed-tool detection,
+partial failure/retry, escaped output, history, stopped-VM startup, per-VM isolation,
+multi-select installation, and mobile layout. Its fixture uses the real API,
+worker threads, and atomic job store with simulated guest execution.
+`tests/test_tool_*.py` cover validation, architecture-specific payloads, checksum
+failure, archive traversal, exact-ID targeting, idempotency, recovery, and shutdown.
+The opt-in real-VM catalog/runtime smoke test is documented in
+[USAGE.md](USAGE.md#18-verification); run it on disposable x86_64 and ARM64 VMs.
 
 When `nft` and `conntrack` are installed, command-parser tests append an invalid
 final command/option so parsing is checked without executing any mutation.

README.md

@@ -310,7 +310,7 @@ and `/dashboard` routes belong to the Nest server.
 
 The Nest dashboard includes a fleet overview, searchable machine collection,
 private network and port-policy view, host diagnostics, and machine provisioning.
-Open a machine to manage its lifecycle, launch terminal sessions, edit or sync
+Open a machine to manage its lifecycle, install developer tools, launch terminal sessions, edit or sync
 its port rules, copy its SSH command, or read recent logs. Select machines in the
 collection for bulk operations.
 Use `Ctrl+K` / `⌘K` (or `/`) to find a machine or page. Fleet data refreshes every
@@ -330,7 +330,7 @@ The server exposes these routes:
 | `GET` | `/diagnostics` | Run host readiness checks for the dashboard/system view |
 | `POST` | `/install` | Install or refresh host assets; body: `{ "force": false }` |
 | `GET` | `/vms` | List VM records and observed status |
-| `POST` | `/vms` | Create; supports `cpu`, `ram`, `storage`, `guest_ip`, `username`, `password`, `name`, `ports`, `no_default_ports` |
+| `POST` | `/vms` | Create; supports `cpu`, `ram`, `storage`, `guest_ip`, `username`, `password`, `name`, `ports`, `no_default_ports`, `tools` |
 | `POST` | `/vms/actions` | Run a fixed snapshot of selected or all VM lifecycle actions |
 | `GET` | `/vms/{identifier}` | Inspect by exact ID, name, or guest IP |
 | `POST` | `/vms/{identifier}/start` | Boot a retained private disk |
@@ -342,10 +342,16 @@ The server exposes these routes:
 | `POST` | `/vms/{vm_id}/ports/sync` | Reapply a managed private port policy for an exact VM ID |
 | `POST` | `/vms/{vm_id}/exec` | Execute a command/script inside a running VM over vsock, with optional root privileges |
 | `POST` | `/vms/{vm_id}/agent/install` | Install/upgrade the execution agent on a stopped VM's private disk |
+| `GET` | `/tools` | Guest tool catalog and supported Python versions |
+| `GET` | `/vms/{vm_id}/tools` | Tool inventory and installation history; `refresh=true` probes the running guest |
+| `POST` | `/vms/{vm_id}/tools/install` | Queue selected guest tools; HTTP 202 with a job ID |
+| `GET` | `/vms/{vm_id}/tools/jobs/{job_id}` | Per-tool progress, results, and bounded installation output |
 | `GET` | `/vms/{identifier}/logs` | Read a bounded, non-following tail of a VM log |
 | `GET` | `/vms/{vm_id}/sessions` | List this VM's saved sessions (Nest) |
 | `POST` | `/vms/{vm_id}/sessions` | Launch a guest terminal; body: `{}` or `{"rows":24,"cols":80}` (Nest) |
 | `GET` | `/vms/{vm_id}/sessions/{session_id}` | Session metadata and status (Nest) |
+| `PATCH` | `/vms/{vm_id}/sessions/{session_id}` | Rename a session; body: `{"name":"Deployment"}` (Nest) |
+| `DELETE` | `/vms/{vm_id}/sessions/{session_id}` | Close the session and delete its saved history; returns 204 (Nest) |
 | `POST` | `/vms/{vm_id}/sessions/{session_id}/close` | End the shell and retain its history (Nest) |
 | `WS` | `/vms/{vm_id}/sessions/{session_id}/terminal` | Authenticated terminal input, output, resize, and replay (Nest) |
 
@@ -363,7 +369,47 @@ Log reads use an optional `tail` query parameter from `0` to `1000`, default
 Bulk requests use an action (`start`, `stop`, `restart`, `delete`, or
 `ports-sync`) plus either `identifiers` or `all_vms: true`; they return a result
 for each fixed target and do not prompt for confirmation. Interactive SSH
-remains CLI-only. The API and dashboard are not background reconcilers.
+remains CLI-only. VM lifecycle changes are explicit; accepted tool installation
+jobs are processed in the background.
+
+### One-click Guest Tools
+
+Choose tools under **New machine → Pick your tools** to install them automatically
+after boot. For an existing VM, open **Machines → select a VM → Tools** and click
+**Install**, or select several tools and click **Install selected**. A stopped VM
+offers **Start & install**. Installation works over vsock with SSH disabled.
+
+The initial catalog targets Ubuntu 22.04 guests on x86_64 and ARM64:
+
+| Tool | Default |
+|---|---|
+| Git, wget, curl, htop | Ubuntu packages |
+| Node.js | 24 LTS (24.21.0), including npm and npx |
+| Python | 3.11, 3.12, and 3.13 with pip and venv; versions can coexist |
+| JDK | OpenJDK 21 LTS, including `java` and `javac` |
+| PHP | 8.5 CLI from the signed Ondřej Surý PHP PPA |
+| Conda | Miniforge 26.5.3-0, using conda-forge |
+| micromamba | 2.9.0 |
+
+Node, Python, Miniforge, and micromamba downloads use pinned release checksums.
+Python uses Astral's standalone distributions, exposed as `python3.11`,
+`python3.12`, and `python3.13`. The guest's OS Python continues serving the agent.
+Downloaded tools live under `/opt/uvm-tools`, with executable links in
+`/usr/local/bin`. Conda environments and caches are writable by the configured
+guest user. **Get started** shows commands, including shell initialization for
+Conda/micromamba activation in Nest's non-login terminals.
+
+Jobs run independently of the browser and show progress and captured output per
+tool. Already available tools are verified and skipped. A failed tool can be
+retried on the created VM. Guest internet access is required for downloads.
+Restart the management server to load the feature; older running VMs may need
+**Restart & install** to apply the current guest-agent setup.
+
+The most recent ten jobs and cached inventory are stored in the VM's private
+`tools.json` file. Queued requests resume when the server starts. An interrupted
+in-flight installation has an uncertain outcome and may finish inside the guest;
+use **Check installed tools** before retrying. See
+[API examples and recovery](USAGE.md#install-guest-tools) for details.
 
 ### MicroVM Terminal Sessions
 
@@ -379,6 +425,13 @@ output. Navigation, browser reloads, and signing out leave the shell running.
 **End session** closes the shell and its processes. Stopping/restarting the VM
 or the Nest server ends live sessions; their saved history remains available.
 
+Use **Rename** in a history row or the terminal toolbar to give any session a
+custom name. Names are trimmed and accept 1–80 printable characters; they persist
+across reloads and server restarts. The session keeps its identity and shell.
+**Delete** removes a session and its saved terminal output after confirmation,
+closing its shell first if it is active. Both actions are scoped to the session's
+microVM.
+
 Nest stores session metadata and the latest **8 MiB of output per session** in
 `$UVM_BASE/nest/sessions.sqlite3` (default `/var/lib/uvm/nest/sessions.sqlite3`).
 The viewer indicates truncated output. Raw keyboard input is not logged

USAGE.md

@@ -823,6 +823,11 @@ passwords in reports.
 5. Use **Session history** to switch between that VM's sessions. Active sessions
    reconnect to their existing shell; ended sessions replay saved output.
 6. Use **End session** when finished, or run `exit` inside the shell.
+7. Use **Rename** in the history row or terminal toolbar to save a custom label.
+   Names accept 1–80 printable characters, with surrounding whitespace trimmed.
+8. Use **Delete** to permanently remove a session and its saved output. Confirming
+   deletion closes an active shell before removing its history. **End session**
+   retains history for later viewing.
 
 Sessions continue while you navigate elsewhere, reload, or sign out of the
 dashboard. VM shutdown/restart or Nest shutdown ends the shell. Session metadata,
@@ -830,6 +835,9 @@ start/end times, exit status, and output survive Nest restarts in
 `$UVM_BASE/nest/sessions.sqlite3`; files are mode `0600` in a mode-`0700` directory.
 Only sessions belonging to the selected exact VM ID are shown. Reusing a VM name
 does not attach another VM's history.
+Renaming works for active and ended sessions and keeps their IDs and URLs stable.
+Deleting the open session clears the terminal selection; other sessions continue
+to appear under their own history entries.
 
 The retained transcript is capped at its latest 8 MiB, and the terminal viewer
 has a 10,000-line scrollback. A truncation note appears when older output is
@@ -912,10 +920,16 @@ curl --cacert /path/to/ca.pem \
 | `POST` | `/vms/{vm_id}/ports/sync` | Reapply the policy for an exact VM ID |
 | `POST` | `/vms/{vm_id}/exec` | Execute a guest command/script over vsock, with optional root privileges |
 | `POST` | `/vms/{vm_id}/agent/install` | Install/upgrade the agent on a stopped VM's disk |
+| `GET` | `/tools` | Authenticated tool catalog and Python version options |
+| `GET` | `/vms/{vm_id}/tools` | Cached tool inventory/history; `refresh=true` checks the running guest |
+| `POST` | `/vms/{vm_id}/tools/install` | Queue tool installation; HTTP 202 with a job ID |
+| `GET` | `/vms/{vm_id}/tools/jobs/{job_id}` | Installation progress, results, and captured output |
 | `GET` | `/vms/{identifier}/logs` | Read a bounded, non-following log tail |
 | `GET` | `/vms/{vm_id}/sessions` | List saved sessions for an exact VM ID (Nest) |
 | `POST` | `/vms/{vm_id}/sessions` | Launch a terminal, with `{}` or optional `rows`/`cols` (Nest) |
 | `GET` | `/vms/{vm_id}/sessions/{session_id}` | Inspect a session (Nest) |
+| `PATCH` | `/vms/{vm_id}/sessions/{session_id}` | Rename with `{"name":"Deployment"}`; returns updated metadata (Nest) |
+| `DELETE` | `/vms/{vm_id}/sessions/{session_id}` | End the shell and remove its record/output; HTTP 204 (Nest) |
 | `POST` | `/vms/{vm_id}/sessions/{session_id}/close` | End a session and retain its output (Nest) |
 | `WS` | `/vms/{vm_id}/sessions/{session_id}/terminal` | Authenticated live terminal/replay (Nest) |
 
@@ -963,7 +977,7 @@ curl -X POST \
 ```
 
 Request fields are `cpu`, `ram`, `storage`, `guest_ip`, `username`, `password`, `name`,
-`ports`, and `no_default_ports`. The name has the same validation as the CLI.
+`ports`, `no_default_ports`, and `tools`. The name has the same validation as the CLI.
 `ports` is an array of objects (or null/omitted), not a CLI colon string.
 Each object has a strict integer `port` from 1 to 65535 and an optional,
 case-sensitive `protocol` of `TCP` (default), `HTTP`, or `HTTPS`.
@@ -978,6 +992,106 @@ The response never includes the password. Omitting credentials still uses
 labels return HTTP `422`. Example passwords above are placeholders; real
 secrets in curl arguments can enter shell history and process listings.
 
+### Install Guest Tools
+
+The dashboard provides tool checkboxes during creation and a **Tools** tab on
+each VM. Choose **Install** for a single tool or **Install selected** for a batch.
+**Start & install** also boots a stopped VM. The catalog covers Ubuntu 22.04 on
+x86_64 and ARM64, with guest internet access for package downloads.
+
+Use the same API selections during creation:
+
+```sh
+curl -X POST \
+  -H "X-UVM-Token: $UVM_API_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"name":"dev-tools","username":"developer","password":"replace-me","tools":[{"id":"git"},{"id":"nodejs"},{"id":"python","version":"3.12"}]}' \
+  "$UVM_API_URL/vms"
+```
+
+The VM response is HTTP 201 after creation. Selected tools install asynchronously;
+read the returned VM's `/tools` endpoint for its job. Omitted `tools` defaults to
+an empty list. An explicit null is rejected.
+
+For an existing VM, use its exact ID from create/list:
+
+```sh
+export VM_ID='<exact-vm-id>'
+curl -H "X-UVM-Token: $UVM_API_TOKEN" "$UVM_API_URL/tools"
+
+curl -X POST \
+  -H "X-UVM-Token: $UVM_API_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"tools":[{"id":"jdk"},{"id":"conda"},{"id":"micromamba"}],"start_if_stopped":true,"request_id":"dev-tools-1"}' \
+  "$UVM_API_URL/vms/$VM_ID/tools/install"
+
+# Use the job ID from the HTTP 202 response.
+export JOB_ID='<tools-job-id>'
+curl -H "X-UVM-Token: $UVM_API_TOKEN" \
+  "$UVM_API_URL/vms/$VM_ID/tools/jobs/$JOB_ID"
+curl -H "X-UVM-Token: $UVM_API_TOKEN" \
+  "$UVM_API_URL/vms/$VM_ID/tools?refresh=true"
+```
+
+Selections are objects with `id` and an optional `version`. IDs are `git`,
+`nodejs`, `python`, `jdk`, `wget`, `curl`, `htop`, `php`, `conda`, and
+`micromamba`. Python requires `version: "3.11"`, `"3.12"`, or `"3.13"`;
+the other tools use the curated catalog default and accept no version override.
+Up to twelve selections are accepted, with identical selections deduplicated.
+Multiple Python versions can be installed together.
+
+Install requests require a nonempty `tools` list. `start_if_stopped` and `restart`
+are strict booleans, both default false. Set `restart: true` to restart first and
+refresh guest-agent setup. In the dashboard, the **Restart & install** recovery
+button appears when agent readiness fails. Unknown tools/options return 422;
+unavailable VM states or a different active install return 409.
+
+`request_id` is an optional idempotency key of 1–80 letters, digits, underscores,
+or hyphens. Reusing it with the same request returns the retained job; different
+selections/options return 409. Use a new key for an intentional retry. Idempotency
+keys last as long as their job in the ten-job history. One batch per VM is active
+at a time, with two VM workers by default and sequential tools within a batch.
+
+Jobs move through `queued`, `starting`, `waiting`, and `installing`, then
+`succeeded`, `failed`, or `interrupted`. Each item reports `installed`,
+`already_installed`, `failed`, `interrupted`, or `skipped`, plus its detected
+version, error, bounded output, and truncation flag. Ordinary recipe failures
+allow subsequent tools to run; uncertain execution outcomes end that batch.
+Output is captured as a 16 KiB tail per tool and returned when the step finishes.
+Default limits are 120 seconds for agent readiness and 1,200 seconds per tool.
+
+`GET /vms/{vm_id}/tools` returns cached inventory, the latest job, and summaries
+of the last ten jobs. `refresh=true` probes all tools in a running VM and updates
+`checked_at`; it does not install software or start a stopped VM. `stale` marks
+old or stopped-VM observations. Installed binaries and their version checks run
+as the configured guest user. Matching installed runtimes are accepted, including
+tools installed outside Nest. The install operation is not a general updater.
+
+Tool jobs survive browser navigation, reload, and logout. The active management
+server owns the queue through `$UVM_BASE/tool-worker.lock`. After server restart,
+queued work resumes; in-flight jobs become interrupted with
+`outcome_unknown: true`. The last command may still finish inside the guest.
+Refresh inventory before retrying; a guest-side lock blocks overlapping installs.
+Stopping/restarting the VM also interrupts a running batch. Successful tools and
+the VM's private disk remain available. Deleting the VM removes its tool history.
+
+Use versioned Python commands, for example `python3.12 -m venv .venv`. For Conda
+or micromamba activation in a Nest terminal, initialize the current POSIX shell:
+
+```sh
+eval "$(conda shell.posix hook)"
+conda create -n myenv python=3.13
+conda activate myenv
+
+# In another shell, for micromamba:
+eval "$(micromamba shell hook --shell posix)"
+micromamba create -n myenv -c conda-forge python=3.13
+micromamba activate myenv
+```
+
+The installer also adds login-shell profiles under `/etc/profile.d`. See
+[the catalog](README.md#one-click-guest-tools) for default versions and sources.
+
 ### Start, Restart, Stop, and Delete
 
 ```sh
@@ -1587,6 +1701,21 @@ exercise real child processes with fake cgroup scopes, local Unix-socket protoco
 fixtures, and temporary ext4 upgrades; they do not prove the real vsock/cgroup
 path inside a VM.
 
+The tools smoke test installs the full catalog in a prepared **disposable Ubuntu
+22.04 VM with a non-root login and no open TCP ports**. It downloads packages,
+checks all runtimes, compiles Java, creates Python venvs, and creates/activates
+temporary Conda and micromamba environments as the guest user. Run it separately
+on x86_64 and ARM64 hosts:
+
+```sh
+UVM_TOOLS_INTEGRATION=1 UVM_TOOLS_VM_ID='<disposable-vm-id>' \
+  python3 -B -m unittest discover -s tests/integration -p test_guest_tools.py -v
+```
+
+Set `UVM_API_URL` and `UVM_API_TOKEN` as above. The test is opt-in and was not run
+against a real microVM during this implementation. Automated API and browser
+checks use simulated guest execution and real job persistence/workers.
+
 ## 19. Get Help
 
 ```sh

nest/dashboard/static/css/sessions.css

@@ -10,7 +10,10 @@
 .session-history > .panel-footnote { margin-top: auto; padding: 15px 8px 0; }
 .session-list { display: grid; gap: 6px; max-height: 510px; overflow-y: auto; }
 .session-list-empty { padding: 24px 9px; color: var(--muted); font-size: 10px; line-height: 1.8; }
-.session-item { display: grid; grid-template-columns: 1fr auto; gap: 7px; padding: 12px 10px; border: 1px solid transparent; border-radius: 12px; color: var(--ink-soft); }
+.session-item { display: grid; gap: 9px; min-width: 0; padding: 12px 10px; border: 1px solid transparent; border-radius: 12px; color: var(--ink-soft); }
+.session-link { display: grid; grid-template-columns: minmax(0, 1fr) auto; gap: 7px; }
+.session-item-actions { display: flex; gap: 15px; padding-top: 8px; border-top: 1px solid var(--line); }
+.session-item-actions .quiet-link { min-height: 24px; font-size: 9px; }
 .session-item:hover { background: #fff8; }
 .session-item.active { background: var(--white); border-color: #d9d0e4; }
 .session-item strong { overflow-wrap: anywhere; font-size: 11px; }
@@ -20,11 +23,14 @@
 .session-state.failed, .session-state.interrupted { background: var(--peach); }
 .terminal-panel { min-width: 0; overflow: hidden; border-radius: 17px; background: #25222d; color: #e8e3ef; box-shadow: var(--shadow-card); }
 .terminal-toolbar { min-height: 56px; display: flex; flex-wrap: wrap; align-items: center; justify-content: space-between; gap: 12px; padding: 12px 16px; border-bottom: 1px solid #ffffff12; background: #302c39; }
-.terminal-toolbar > div:first-child { display: flex; flex-wrap: wrap; align-items: center; gap: 10px; }
-.terminal-toolbar strong { font-size: 11px; }
+.terminal-toolbar > div:first-child { display: flex; flex-wrap: wrap; align-items: center; gap: 10px; min-width: 0; }
+.terminal-toolbar strong { font-size: 11px; overflow-wrap: anywhere; }
 .terminal-status { color: #b8afc6; font-size: 9px; }
 .terminal-toolbar .button { color: #e8e3ef; background: #ffffff0c; border-color: #ffffff17; }
 .terminal-toolbar .button:hover { background: #ffffff19; }
+#session-delete { color: #e5b4aa; }
+.session-rename-dialog .field { margin-top: 22px; }
+#session-rename-error { color: var(--red); }
 .terminal-lights { display: inline-flex; gap: 4px; margin-right: 3px; }
 .terminal-lights i { height: 6px; width: 6px; border-radius: 50%; background: var(--peach); }
 .terminal-lights i:nth-child(2) { background: var(--yellow); }

nest/dashboard/static/css/tools.css

@@ -0,0 +1,60 @@
+.tool-choices { display: grid; grid-template-columns: repeat(2, minmax(0, 1fr)); gap: 9px; }
+.tool-choice { display: flex; align-items: center; gap: 11px; padding: 12px; border: 1px solid var(--line); border-radius: 13px; background: #ffffff75; cursor: pointer; }
+.tool-choice:hover { background: #fff; }
+.tool-choice.is-selected { background: var(--yellow-soft); border-color: #ded09a; }
+.tool-choice > span { display: grid; gap: 2px; min-width: 0; }
+.tool-choice strong { font-size: 11px; }
+.tool-choice small { color: var(--muted); font-size: 9px; }
+.tool-choice input, .tool-card input { flex: 0 0 auto; }
+.tools-section-heading { margin-block: 7px 15px; }
+.tools-section-heading .button-row { justify-content: flex-end; }
+.tools-section-heading .button-row > span { font-size: 10px; }
+.tools-note { min-height: 22px; margin-bottom: 12px; }
+.tools-error { margin-bottom: 12px; }
+.tools-error:empty { display: none; }
+.tools-grid { display: grid; grid-template-columns: repeat(3, minmax(0, 1fr)); gap: 14px; }
+.tool-card { display: flex; flex-direction: column; min-width: 0; padding: 18px; border-radius: 18px; border: 1px solid #ffffffbd; background: #faf9fcbf; box-shadow: var(--shadow-card); }
+.tool-card-heading { display: flex; gap: 11px; align-items: center; cursor: pointer; }
+.tool-card-heading > span { min-width: 0; display: grid; gap: 4px; }
+.tool-card-heading strong { font-size: 14px; letter-spacing: -.02em; }
+.tool-card-heading small { font-size: 9px; color: var(--muted); }
+.tool-card > p { margin: 12px 0 5px; color: var(--muted); font-size: 10px; }
+.tool-source { font-size: 8px; color: var(--faint); margin-bottom: 17px; }
+.tool-card-footer { display: flex; align-items: center; justify-content: space-between; flex-wrap: wrap; gap: 8px; margin-top: auto; }
+.tool-installed { border-color: #b8c6ae70; }
+.tool-installed button:disabled { cursor: default; background: var(--sage-soft); color: #64705c; opacity: 1; }
+.tool-state { font-size: 9px; color: var(--muted); }
+.tool-state-good { color: #68765c; }
+.tool-state-error { color: #a45f57; }
+.tool-state-active { color: #827037; }
+.tool-state-active::before { content: ''; display: inline-block; width: 5px; height: 5px; margin-right: 5px; border-radius: 50%; background: var(--yellow); animation: tool-pulse 1.5s ease-in-out infinite; }
+.tool-usage { margin-top: 13px; padding-top: 10px; border-top: 1px solid var(--line); font-size: 9px; color: var(--muted); }
+.tool-usage summary { cursor: pointer; }
+.tool-usage code { display: block; margin: 9px 0; font-family: var(--mono); font-size: 9px; overflow-wrap: anywhere; }
+.tool-job { margin-top: 22px; padding: 22px; }
+.tool-job .section-heading { flex-wrap: wrap; gap: 10px; margin-bottom: 8px; }
+.tool-job .select-wrap { max-width: 100%; }
+.tool-job select { max-width: 100%; }
+.tool-job-message { color: var(--muted); font-size: 11px; overflow-wrap: anywhere; }
+.tool-progress { display: block; width: 100%; height: 5px; margin: 15px 0; border: none; border-radius: 10px; background: #e7e2eb; accent-color: #b8ad69; }
+.tool-progress::-webkit-progress-bar { background: #e7e2eb; border-radius: 10px; }
+.tool-progress::-webkit-progress-value { background: #c3b875; border-radius: 10px; }
+.tool-progress::-moz-progress-bar { background: #c3b875; border-radius: 10px; }
+.tool-job-items { padding: 0; margin: 0; list-style: none; }
+.tool-job-items > li { padding: 10px 0; border-top: 1px solid var(--line); }
+.tool-job-items summary { cursor: pointer; font-size: 11px; }
+.tool-job-items summary > .tool-state { float: right; margin-left: 12px; }
+.tool-job-items .form-error { margin-top: 10px; }
+.tool-output { max-height: 240px; overflow: auto; margin: 10px 0 0; padding: 15px; border-radius: 12px; background: #34313c; color: #ede8f1; white-space: pre-wrap; overflow-wrap: anywhere; font: 10px/1.7 var(--mono); }
+.tools-recovery { margin-top: 16px; }
+@keyframes tool-pulse { 50% { opacity: .35; } }
+@media (max-width: 1250px) { .tools-grid { grid-template-columns: repeat(2, minmax(0, 1fr)); } }
+@media (max-width: 700px) {
+  .tools-section-heading { align-items: flex-start; flex-direction: column; }
+  .tools-section-heading .button-row { justify-content: flex-start; }
+  .tools-grid { grid-template-columns: 1fr; }
+  .tool-card { padding: 16px; }
+  .tool-job { padding: 16px; }
+  .tool-choice { padding: 10px; gap: 8px; }
+}
+@media (prefers-reduced-motion: reduce) { .tool-state-active::before { animation: none; } }

nest/dashboard/static/index.html

@@ -11,11 +11,13 @@
   <link rel="preload" href="/dashboard/static/fonts/manrope-latin.woff2" as="font" type="font/woff2" crossorigin />
   <link rel="stylesheet" href="/dashboard/static/vendor/xterm/css/xterm.css?v=5.5.0" />
   <link rel="stylesheet" href="/dashboard/static/css/dashboard.css?v=3" />
-  <link rel="stylesheet" href="/dashboard/static/css/sessions.css?v=1" />
+  <link rel="stylesheet" href="/dashboard/static/css/sessions.css?v=2" />
+  <link rel="stylesheet" href="/dashboard/static/css/tools.css?v=1" />
   <script src="/dashboard/static/vendor/xterm/lib/xterm.js?v=5.5.0" defer></script>
   <script src="/dashboard/static/vendor/xterm-fit/lib/addon-fit.js?v=0.10.0" defer></script>
-  <script src="/dashboard/static/js/sessions.js?v=2" defer></script>
-  <script src="/dashboard/static/js/dashboard.js?v=4" defer></script>
+  <script src="/dashboard/static/js/sessions.js?v=3" defer></script>
+  <script src="/dashboard/static/js/tools.js?v=1" defer></script>
+  <script src="/dashboard/static/js/dashboard.js?v=6" defer></script>
 </head>
 <body>
   <svg class="icon-sprite" xmlns="http://www.w3.org/2000/svg" aria-hidden="true">
@@ -178,6 +180,7 @@
             <div id="detail-error" class="notice notice-error" hidden></div>
             <nav class="machine-tabs" aria-label="Selected machine views">
               <a id="vm-overview-tab" class="machine-tab" href="#vms">Overview</a>
+              <a id="vm-tools-tab" class="machine-tab" href="#vms"><svg class="icon"><use href="#i-download"/></svg>Tools</a>
               <a id="vm-sessions-tab" class="machine-tab" href="#vms"><svg class="icon"><use href="#i-terminal"/></svg>Sessions <span id="vm-session-count" class="nav-count">—</span></a>
             </nav>
             <div id="vm-overview-panel">
@@ -197,6 +200,19 @@
             <article class="soft-panel log-panel"><header class="section-heading"><div><h2>A window into the runtime</h2><span>Recent Firecracker log entries.</span></div><div class="button-row"><label class="select-wrap"><span class="sr-only">Log tail length</span><select id="log-tail"><option value="100">100 lines</option><option value="250">250 lines</option><option value="500">500 lines</option><option value="1000">1,000 lines</option></select></label><button id="log-refresh-btn" class="icon-button" type="button" data-action="refresh-logs" aria-label="Refresh logs"><svg class="icon"><use href="#i-refresh"/></svg></button></div></header><pre id="log-output" class="log-output" tabindex="0"></pre><p id="log-meta" class="panel-footnote"></p></article>
             <div class="detail-delete-row"><span>Every machine has its own private disk.</span><button class="quiet-link danger-text" type="button" data-action="delete-current"><svg class="icon"><use href="#i-trash"/></svg>Delete machine</button></div>
             </div>
+            <section id="vm-tools-panel" aria-label="Selected machine tools" hidden>
+              <header class="section-heading tools-section-heading"><div><h2>Tools for your next idea</h2><span>One click. Ready inside this machine.</span></div><div class="button-row"><span id="tools-selected-count" class="muted">0 selected</span><button id="tools-install-selected" class="button button-primary" type="button" disabled>Install selected</button><button id="tools-refresh" class="icon-button" type="button" aria-label="Check installed tools" title="Check installed tools"><svg class="icon"><use href="#i-refresh"/></svg></button></div></header>
+              <p id="tools-note" class="panel-footnote tools-note" role="status"></p>
+              <p id="tools-error" class="form-error tools-error" role="alert"></p>
+              <div id="tools-list" class="tools-grid"></div>
+              <article id="tools-job" class="soft-panel tool-job" hidden>
+                <header class="section-heading"><h3 id="tools-job-title">Installation</h3><label class="select-wrap"><span class="sr-only">Installation history</span><select id="tools-job-select"></select></label></header>
+                <p id="tools-job-message" class="tool-job-message" role="status"></p>
+                <progress id="tools-progress" class="tool-progress" max="1" value="0" aria-label="Installation progress"></progress>
+                <ol id="tools-job-items" class="tool-job-items"></ol>
+              </article>
+              <div class="button-row tools-recovery"><button id="tools-retry" class="button button-small" type="button" hidden>Retry unfinished tools</button><button id="tools-restart" class="button button-small" type="button" hidden>Restart &amp; install</button></div>
+            </section>
             <section id="vm-sessions-panel" aria-label="Selected machine sessions" hidden>
               <header class="section-heading session-section-heading"><div><h2>Sessions</h2><span>A shell for each task. A history for this machine.</span></div><button class="button button-primary" type="button" data-action="launch-session"><svg class="icon"><use href="#i-plus"/></svg><span>Launch session</span></button></header>
               <p id="sessions-error" class="form-error" role="alert"></p>
@@ -207,7 +223,7 @@
                   <p class="panel-footnote">Saved for this microVM.</p>
                 </aside>
                 <article class="terminal-panel">
-                  <header class="terminal-toolbar"><div><span class="terminal-lights" aria-hidden="true"><i></i><i></i><i></i></span><strong id="session-title">Terminal</strong><span id="session-status" class="terminal-status">Select a session</span></div><div class="button-row"><button id="session-reconnect" class="button button-small" type="button" hidden>Reconnect</button><button id="session-end" class="button button-small" type="button" disabled>End session</button></div></header>
+                  <header class="terminal-toolbar"><div><span class="terminal-lights" aria-hidden="true"><i></i><i></i><i></i></span><strong id="session-title">Terminal</strong><span id="session-status" class="terminal-status">Select a session</span></div><div class="button-row"><button id="session-reconnect" class="button button-small" type="button" hidden>Reconnect</button><button id="session-rename" class="button button-small" type="button" disabled>Rename</button><button id="session-end" class="button button-small" type="button" disabled>End session</button><button id="session-delete" class="button button-small" type="button" disabled>Delete</button></div></header>
                   <div id="session-blank" class="terminal-blank"><svg class="icon"><use href="#i-terminal"/></svg><h3>A little room to work.</h3><p>Select a saved session, or launch a fresh terminal<br />inside this microVM.</p><button class="button button-primary" type="button" data-action="launch-session"><svg class="icon"><use href="#i-plus"/></svg><span>Launch session</span></button></div>
                   <div id="session-terminal" class="session-terminal" aria-label="MicroVM terminal" hidden></div>
                   <footer id="session-meta" class="terminal-footer">Your sessions and terminal output are saved on the host.</footer>
@@ -229,7 +245,8 @@
             <div class="create-fields"><section class="soft-panel form-section"><header class="section-heading"><div><h2><span class="step-number">01</span>A name and a little space</h2><span>Give your new machine a shape.</span></div></header><label class="field" for="create-name"><span>Machine name <small>Optional</small></span><input id="create-name" type="text" name="name" placeholder="e.g. my-little-server" pattern="[a-z][a-z0-9\-]{0,62}" maxlength="63" autocomplete="off" /></label><fieldset class="preset-fieldset"><legend>Start with a size</legend><div class="size-presets"><button class="size-preset active" type="button" data-preset="small" aria-pressed="true"><svg class="icon"><use href="#i-server"/></svg><strong>Small</strong><span>1 vCPU · 512 MiB</span></button><button class="size-preset" type="button" data-preset="balanced" aria-pressed="false"><svg class="icon"><use href="#i-server"/></svg><strong>Balanced</strong><span>2 vCPU · 1 GiB</span></button><button class="size-preset" type="button" data-preset="roomy" aria-pressed="false"><svg class="icon"><use href="#i-server"/></svg><strong>Roomy</strong><span>4 vCPU · 2 GiB</span></button></div></fieldset><div class="form-grid"><label class="field" for="create-cpu"><span>CPU capacity</span><input id="create-cpu" name="cpu" type="number" value="1" min="0.01" step="any" required /><small>Fractional requests round up to whole vCPUs.</small></label><label class="field" for="create-ram"><span>Memory</span><input id="create-ram" name="ram" type="text" value="512M" required /><small>At least 128 MiB. Try 512M, 1G, or 2G.</small></label></div></section>
             <section class="soft-panel form-section"><label class="field" for="create-storage"><span>Private storage</span><input id="create-storage" name="storage" type="text" value="30G" required /><small>Minimum 30 GiB. Try 30G, 50G, or 1T. You can resize this disk later.</small></label></section>
             <section class="soft-panel form-section"><header class="section-heading"><div><h2><span class="step-number">02</span>A way to connect</h2><span>Private networking and guest access.</span></div></header><label class="field" for="create-guest-ip"><span>Guest IPv4 address <small>Optional</small></span><input id="create-guest-ip" name="guest_ip" type="text" placeholder="Automatically assigned" autocomplete="off" /></label><div class="form-grid"><label class="field" for="create-username"><span>Guest user</span><input id="create-username" name="username" type="text" value="root" autocomplete="username" required /></label><label class="field" for="create-password"><span>Guest password</span><span class="password-field"><input id="create-password" name="password" type="password" value="root" autocomplete="new-password" maxlength="128" required /><button class="icon-button" type="button" data-action="toggle-password" aria-label="Show password" aria-pressed="false"><svg class="icon"><use href="#i-eye"/></svg></button></span></label></div><p class="panel-footnote">Missing users are created with a home directory and sudo access using their guest password. Credentials are applied to the machine’s private disk.</p></section>
-            <section class="soft-panel form-section"><header class="section-heading"><div><h2><span class="step-number">03</span>Open the right doors</h2><span>Choose which private TCP services are allowed.</span></div></header><label class="toggle-field"><span><strong>Include the default SSH rule</strong><small>Allow TCP port 22 for terminal access.</small></span><input id="create-default-ssh" name="default_ssh" type="checkbox" checked /><span class="toggle-track" aria-hidden="true"></span></label><label class="field" for="create-ports"><span>Additional port rules <small>Optional</small></span><input id="create-ports" name="ports" type="text" placeholder="80:HTTP, 443:HTTPS" spellcheck="false" /><small>Separate rules with commas. HTTP and HTTPS both use TCP.</small></label></section></div>
+            <section class="soft-panel form-section"><header class="section-heading"><div><h2><span class="step-number">03</span>Open the right doors</h2><span>Choose which private TCP services are allowed.</span></div></header><label class="toggle-field"><span><strong>Include the default SSH rule</strong><small>Allow TCP port 22 for terminal access.</small></span><input id="create-default-ssh" name="default_ssh" type="checkbox" checked /><span class="toggle-track" aria-hidden="true"></span></label><label class="field" for="create-ports"><span>Additional port rules <small>Optional</small></span><input id="create-ports" name="ports" type="text" placeholder="80:HTTP, 443:HTTPS" spellcheck="false" /><small>Separate rules with commas. HTTP and HTTPS both use TCP.</small></label></section>
+            <section class="soft-panel form-section create-tools-section"><header class="section-heading"><div><h2><span class="step-number">04</span>Pick your tools</h2><span>Optional. Install automatically after your machine boots.</span></div><button id="create-tools-refresh" class="icon-button" type="button" aria-label="Reload tool catalog"><svg class="icon"><use href="#i-refresh"/></svg></button></header><div id="create-tools-list" class="tool-choices"><span class="muted">Loading available tools…</span></div><p id="create-tools-error" class="form-error" role="alert"></p><p class="panel-footnote">Python versions can coexist. You can add more tools later from the machine’s Tools tab.</p></section></div>
             <aside class="create-summary soft-panel"><div class="summary-art" aria-hidden="true"><div></div><div></div><span><svg class="icon"><use href="#i-server"/></svg></span></div><span class="subtle-label">Your next little machine</span><h2 id="create-summary-name">Room for an idea.</h2><p>A private disk, a network identity,<br />and a fresh start.</p><dl id="create-summary-details" class="detail-list"></dl><p id="create-error" class="form-error" role="alert"></p><button id="create-submit" class="button button-primary button-wide" type="submit"><span>Create machine</span><svg class="icon"><use href="#i-arrow"/></svg></button><a href="#vms" class="quiet-link summary-cancel">Maybe later</a><span class="summary-note"><svg class="icon"><use href="#i-spark"/></svg>Built on Firecracker</span></aside>
           </form>
         </section>
@@ -246,6 +263,15 @@
   <div id="action-menu" class="action-menu" role="menu" aria-label="Machine actions" hidden></div>
   <div id="toast-container" class="toast-container" aria-live="polite" aria-atomic="false"></div>
   <dialog id="confirm-dialog" class="confirm-dialog" aria-labelledby="confirm-title" aria-describedby="confirm-body"><form method="dialog"><span class="round-icon" id="confirm-icon"><svg class="icon"><use href="#i-info"/></svg></span><span class="subtle-label">A moment before we continue</span><h2 id="confirm-title"></h2><p id="confirm-body"></p><div class="dialog-actions"><button class="button button-quiet" value="cancel" autofocus>Cancel</button><button id="confirm-ok" class="button button-primary" value="confirm">Confirm</button></div></form></dialog>
+  <dialog id="session-rename-dialog" class="confirm-dialog session-rename-dialog" aria-labelledby="session-rename-title">
+    <form id="session-rename-form">
+      <span class="subtle-label">A name for this workspace</span>
+      <h2 id="session-rename-title">Rename session</h2>
+      <label class="field" for="session-name-input"><span>Session name</span><input id="session-name-input" type="text" maxlength="80" autocomplete="off" spellcheck="false" required aria-describedby="session-rename-error" /></label>
+      <p id="session-rename-error" class="form-error" role="alert"></p>
+      <div class="dialog-actions"><button id="session-rename-cancel" class="button button-quiet" type="button">Cancel</button><button id="session-rename-save" class="button button-primary" type="submit">Save name</button></div>
+    </form>
+  </dialog>
   <noscript><p class="noscript-message">Enable JavaScript to connect to your Nest workspace.</p></noscript>
 </body>
 </html>

nest/dashboard/static/js/dashboard.js

@@ -53,6 +53,11 @@
     onExpired: () => logout('Your token was not accepted. Connect again to continue.'),
     esc, formatDate, toast, confirmAction,
   });
+  const tools = new window.NestTools({
+    api, esc, toast, formatDate,
+    onSelectionChange: updateCreateSummary,
+    onMutation: refreshAfterMutation,
+  });
 
   function esc(value) {
     return String(value ?? '').replace(/[&<>"']/g, (character) => ({
@@ -170,7 +175,7 @@
     return `The server returned HTTP ${status}.`;
   }
 
-  async function api(path, { method = 'GET', body } = {}) {
+  async function api(path, { method = 'GET', body, timeout = 20000 } = {}) {
     const session = state.session;
     const controller = new AbortController();
     state.controllers.add(controller);
@@ -179,7 +184,7 @@
     const timer = method === 'GET' ? window.setTimeout(() => {
       timedOut = true;
       controller.abort();
-    }, 20000) : null;
+    }, timeout) : null;
     try {
       const response = await fetch(path, {
         method,
@@ -612,17 +617,21 @@
       : actionButton('start', 'Start machine', 'play', true))
       + `<button class="button button-primary" type="button" data-action="launch-session">${icon('terminal')}<span>Launch session</span></button>`);
     const showingSessions = state.detailTab === 'sessions';
-    $('#vm-overview-panel').hidden = showingSessions;
+    const showingTools = state.detailTab === 'tools';
+    $('#vm-overview-panel').hidden = showingSessions || showingTools;
     $('#vm-sessions-panel').hidden = !showingSessions;
+    $('#vm-tools-panel').hidden = !showingTools;
     $('#vm-overview-tab').href = vmHref(vm.id);
     $('#vm-sessions-tab').href = `${vmHref(vm.id)}/sessions`;
-    [['#vm-overview-tab', !showingSessions], ['#vm-sessions-tab', showingSessions]].forEach(([selector, active]) => {
+    $('#vm-tools-tab').href = `${vmHref(vm.id)}/tools`;
+    [['#vm-overview-tab', !showingSessions && !showingTools], ['#vm-sessions-tab', showingSessions], ['#vm-tools-tab', showingTools]].forEach(([selector, active]) => {
       $(selector).classList.toggle('active', active);
       if (active) $(selector).setAttribute('aria-current', 'page');
       else $(selector).removeAttribute('aria-current');
     });
-    text('#breadcrumb', `Machines / ${machineName(vm)}${showingSessions ? ' / Sessions' : ''}`);
+    text('#breadcrumb', `Machines / ${machineName(vm)}${showingSessions ? ' / Sessions' : showingTools ? ' / Tools' : ''}`);
     sessions.sync(vm, state.terminalId, showingSessions);
+    tools.sync(vm, showingTools);
     html('#detail-config', detailRows([
       ['Recorded state', label(vm.status)], ['Observed state', label(runtimeStatus(vm))],
       ['vCPUs', vcpu(vm)], ['Memory', formatRam(vm.ram_mib)], ['Storage', formatStorage(vm.storage_mib)], ['Guest user', vm.username],
@@ -685,6 +694,7 @@
         $('#detail-error').hidden = false;
       } else {
         sessions.leave();
+        tools.leave();
         state.detail = null;
         $('#detail-content').hidden = true;
         $('#detail-loading').hidden = false;
@@ -927,6 +937,7 @@
       ['Compute', Number.isFinite(cpu) && cpu > 0 ? `${Math.ceil(cpu)} vCPU` : '—'],
       ['Memory', memory], ['Storage', storage], ['Guest IP', $('#create-guest-ip').value.trim() || 'Automatic'],
       ['Guest user', $('#create-username').value.trim() || '—'], ['Allowed ports', ports],
+      ['Tools', tools.createSummary()],
     ]));
     const presets = { small: [1, 512], balanced: [2, 1024], roomy: [4, 2048] };
     let mib = null;
@@ -966,6 +977,7 @@
         password: $('#create-password').value,
         ports: parsePorts($('#create-ports').value),
         no_default_ports: !$('#create-default-ssh').checked,
+        tools: tools.createSelections(),
       };
       if (!payload.password) throw new Error('Add a password for the guest account.');
     } catch (error) {
@@ -977,22 +989,25 @@
     const previous = button.innerHTML;
     const unlock = lockForm($('#create-form'));
     state.busy.add('create');
+    tools.setCreateBusy(true);
     button.textContent = 'Creating your machine…';
     try {
       const vm = await api('/vms', { method: 'POST', body: payload });
       if (session !== state.session) return;
       if (!vm?.id) throw new Error('The machine response was not recognized. Refresh the registry to check its state.');
       $('#create-form').reset();
+      tools.resetCreate();
       resetPasswordVisibility();
       updateCreateSummary();
-      toast('A new little machine, ready for you.');
-      location.hash = vmHref(vm.id);
+      toast(payload.tools.length ? 'Machine created. Your selected tools are queued for installation.' : 'A new little machine, ready for you.');
+      location.hash = vmHref(vm.id) + (payload.tools.length ? '/tools' : '');
       await loadVms();
     } catch (error) {
       if (!isAborted(error) && session === state.session) text('#create-error', error.message);
     } finally {
       unlock();
       button.innerHTML = previous;
+      tools.setCreateBusy(false);
       if (session === state.session) state.busy.delete('create');
     }
   }
@@ -1151,6 +1166,8 @@
         if (parts[2] === 'sessions' && parts.length <= 4) {
           detailTab = 'sessions';
           terminalId = parts[3] ? decodeURIComponent(parts[3]) : null;
+        } else if (parts[2] === 'tools' && parts.length === 3) {
+          detailTab = 'tools';
         } else if (parts.length !== 2) identifier = null;
       } catch { identifier = null; }
       value = identifier ? 'vm-detail' : 'overview';
@@ -1164,6 +1181,7 @@
     state.detailTab = detailTab;
     state.terminalId = terminalId;
     sessions.leave();
+    tools.leave();
     state.detail = null;
     state.portEditing = false;
     state.resizeEditing = false;
@@ -1192,7 +1210,7 @@
       loadDetail(identifier);
     }
     if (value === 'system') loadDiagnostics();
-    if (value === 'create') updateCreateSummary();
+    if (value === 'create') { updateCreateSummary(); tools.mountCreate(); }
     if (!initial && value !== 'vm-detail') $(`#page-${value} h1`)?.focus({ preventScroll: true });
   }
 
@@ -1253,6 +1271,7 @@
 
   function logout(message = '') {
     sessions.leave();
+    tools.reset();
     state.session += 1;
     state.controllers.forEach((controller) => controller.abort());
     state.controllers.clear();
@@ -1428,7 +1447,7 @@
       if (event.target.closest('.xterm')) return;
       const editing = event.target.closest('input, textarea, select, [contenteditable="true"]');
       if (state.authenticated && ((event.key.toLowerCase() === 'k' && (event.metaKey || event.ctrlKey)) || (event.key === '/' && !editing))) {
-        if ($('#confirm-dialog').open) return;
+        if (document.querySelector('dialog[open]')) return;
         event.preventDefault();
         focusSearch();
       }

nest/dashboard/static/js/sessions.js

@@ -19,12 +19,31 @@
       this.generation = 0;
       this.launching = new Set();
       this.upgrading = new Set();
+      this.mutating = new Map();
       this.offset = 0;
       this.renderPending = Promise.resolve();
       $('#sessions-refresh').addEventListener('click', () => this.refresh());
       $('#session-end').addEventListener('click', () => this.end());
       $('#session-reconnect').addEventListener('click', () => this.connectSocket());
       $('#session-install-agent').addEventListener('click', () => this.installAgent());
+      $('#session-rename').addEventListener('click', () => this.rename(this.current));
+      $('#session-delete').addEventListener('click', () => this.deleteSession(this.current));
+      $('#session-list').addEventListener('click', (event) => {
+        const button = event.target.closest('[data-session-action]');
+        if (!button || button.disabled) return;
+        const record = this.records.find((item) => item.id === button.dataset.sessionId);
+        if (button.dataset.sessionAction === 'rename') this.rename(record);
+        if (button.dataset.sessionAction === 'delete') this.deleteSession(record);
+      });
+      $('#session-rename-form').addEventListener('submit', (event) => { event.preventDefault(); this.saveName(); });
+      $('#session-rename-cancel').addEventListener('click', () => $('#session-rename-dialog').close());
+      $('#session-rename-dialog').addEventListener('close', () => {
+        if (!$('#session-rename-dialog').open) this.renameTarget = null;
+      });
+      $('#session-rename-dialog').addEventListener('cancel', (event) => {
+        if (this.isBusy(this.renameTarget?.id)) event.preventDefault();
+      });
+      $('#session-name-input').addEventListener('input', () => { $('#session-rename-error').textContent = ''; });
     }
 
     sync(vm, selectedId, visible) {
@@ -73,10 +92,16 @@
 
     renderList() {
       const markup = this.records.length ? this.records.map((record) => `
-        <a class="session-item${record.id === this.selectedId ? ' active' : ''}" href="${href(this.vm.id, record.id)}" ${record.id === this.selectedId ? 'aria-current="page"' : ''}>
+        <div class="session-item${record.id === this.selectedId ? ' active' : ''}" data-session-id="${this.esc(record.id)}">
+          <a class="session-link" href="${href(this.vm.id, record.id)}" ${record.id === this.selectedId ? 'aria-current="page"' : ''}>
             <strong>${this.esc(record.name)}</strong><i class="session-state ${this.esc(record.status)}" aria-hidden="true"></i>
             <small>${this.esc(this.formatDate(record.created_at))}<br />${this.esc(record.status)}${record.exit_code != null ? ` · exit ${record.exit_code}` : ''}</small>
-        </a>`).join('') : `<p class="session-list-empty">${this.loading ? 'Loading sessions…' : 'No sessions yet.<br />Launch a session to get started.'}</p>`;
+          </a>
+          <div class="session-item-actions">
+            <button class="quiet-link" type="button" data-session-action="rename" data-session-id="${this.esc(record.id)}" aria-label="Rename ${this.esc(record.name)}" ${this.isBusy(record.id) ? 'disabled' : ''}>Rename</button>
+            <button class="quiet-link danger-text" type="button" data-session-action="delete" data-session-id="${this.esc(record.id)}" aria-label="Delete ${this.esc(record.name)}" ${this.isBusy(record.id) ? 'disabled' : ''}>${this.mutating.get(record.id) === 'delete' ? 'Deleting…' : 'Delete'}</button>
+          </div>
+        </div>`).join('') : `<p class="session-list-empty">${this.loading ? 'Loading sessions…' : 'No sessions yet.<br />Launch a session to get started.'}</p>`;
       if ($('#session-list').innerHTML !== markup) $('#session-list').innerHTML = markup;
     }
 
@@ -93,6 +118,12 @@
         this.records = records;
         if (this.current) {
           const latest = records.find((record) => record.id === this.current.id);
+          if (!latest) {
+            const localDelete = this.mutating.get(this.current.id) === 'delete';
+            this.removeRecord(vmId, this.current.id);
+            if (!localDelete) this.toast('This session was deleted.');
+            return;
+          }
           if (latest && latest.updated_at >= this.current.updated_at) this.current = latest;
           else if (latest) this.records[this.records.indexOf(latest)] = this.current;
         }
@@ -204,7 +235,8 @@
     }
 
     canType() {
-      return this.socket?.readyState === WebSocket.OPEN && this.current?.status === 'active' && this.caughtUp;
+      return this.socket?.readyState === WebSocket.OPEN && this.current?.status === 'active' && this.caughtUp
+        && this.ending !== this.current.id && this.mutating.get(this.current.id) !== 'delete';
     }
 
     sendSize(rows, cols) {
@@ -239,7 +271,7 @@
           const message = JSON.parse(event.data);
           if (message.type === 'session') {
             const wasReady = this.caughtUp;
-            this.current = message.session;
+            if (message.session.updated_at >= this.current.updated_at) this.current = message.session;
             if (message.offset > this.offset) {
               this.term.reset();
               this.offset = message.offset;
@@ -248,7 +280,7 @@
             this.caughtUp = this.offset >= this.replayEnd;
             this.connection = this.caughtUp ? 'Connected' : 'Loading history…';
             const index = this.records.findIndex((record) => record.id === message.session.id);
-            if (index >= 0) this.records[index] = message.session;
+            if (index >= 0) this.records[index] = this.current;
             this.renderList();
             this.renderStatus();
             if (!wasReady && this.canType()) { this.sendSize(this.term.rows, this.term.cols); this.term.focus(); }
@@ -271,6 +303,12 @@
                 resolve();
               });
             });
+          } else if (message.type === 'deleted') {
+            const vmId = this.vm.id;
+            const sessionId = this.current.id;
+            const localDelete = this.mutating.get(sessionId) === 'delete';
+            this.removeRecord(vmId, sessionId);
+            if (!localDelete) this.toast('This session was deleted.');
           } else if (message.type === 'error') {
             rejected = true;
             this.showError(message.message);
@@ -298,7 +336,10 @@
       const active = record?.status === 'active';
       $('#session-title').textContent = record ? `${record.name} · ${this.vm?.name || this.vm?.id || ''}` : 'Terminal';
       $('#session-status').textContent = record ? active ? this.connection || 'Connecting…' : `${record.status}${record.exit_code != null ? ` · exit ${record.exit_code}` : ''}` : 'Select a session';
-      $('#session-end').disabled = !active || this.ending === record?.id;
+      $('#session-end').disabled = !active || this.isBusy(record?.id);
+      $('#session-rename').disabled = !record || this.isBusy(record.id);
+      $('#session-delete').disabled = !record || this.isBusy(record.id);
+      $('#session-delete').textContent = record && this.mutating.get(record.id) === 'delete' ? 'Deleting…' : 'Delete';
       $('#session-reconnect').hidden = !active || this.socket?.readyState === WebSocket.OPEN;
       if (this.term) this.term.options.disableStdin = !this.canType();
       $('#session-meta').textContent = record
@@ -308,10 +349,11 @@
 
     async end() {
       const record = this.current;
-      if (!record || this.ending) return;
+      if (!record || this.isBusy(record.id)) return;
       const confirmed = await this.confirmAction(`End ${record.name}?`, 'Close this shell and its running processes. Its saved terminal output will be available in session history.', 'End session');
       if (!confirmed) return;
       this.ending = record.id;
+      this.renderList();
       this.renderStatus();
       try {
         const ended = await this.api(path(record.vm_id, `/${encodeURIComponent(record.id)}/close`), { method: 'POST' });
@@ -321,10 +363,124 @@
         if (error.name !== 'AbortError' && this.vm?.id === record.vm_id) this.showError(error.message);
       } finally {
         this.ending = null;
+        this.renderList();
         this.renderStatus();
       }
     }
 
+    isBusy(id) {
+      return Boolean(id && (this.mutating.has(id) || this.ending === id));
+    }
+
+    invalidateRefresh() {
+      this.sequence += 1;
+      this.loading = false;
+      this.lastRefresh = 0;
+      $('#sessions-refresh').disabled = false;
+    }
+
+    rename(record) {
+      if (!record || this.isBusy(record.id) || document.querySelector('dialog[open]')) return;
+      this.renameTarget = record;
+      $('#session-name-input').value = record.name;
+      $('#session-name-input').disabled = false;
+      $('#session-rename-save').disabled = false;
+      $('#session-rename-save').textContent = 'Save name';
+      $('#session-rename-cancel').disabled = false;
+      $('#session-rename-error').textContent = '';
+      $('#session-rename-dialog').showModal();
+      $('#session-name-input').focus();
+      $('#session-name-input').select();
+    }
+
+    async saveName() {
+      const target = this.renameTarget;
+      if (!target || this.isBusy(target.id)) return;
+      const name = $('#session-name-input').value.trim();
+      if (!name) {
+        $('#session-rename-error').textContent = 'Enter a session name.';
+        $('#session-name-input').focus();
+        return;
+      }
+      const token = this.getToken();
+      this.mutating.set(target.id, 'rename');
+      $('#session-name-input').disabled = true;
+      $('#session-rename-save').disabled = true;
+      $('#session-rename-save').textContent = 'Saving…';
+      $('#session-rename-cancel').disabled = true;
+      this.renderList();
+      this.renderStatus();
+      try {
+        const renamed = await this.api(path(target.vm_id, `/${encodeURIComponent(target.id)}`), { method: 'PATCH', body: { name } });
+        if (this.renameTarget === target) $('#session-rename-dialog').close();
+        if (token === this.getToken() && this.vm?.id === target.vm_id) {
+          this.invalidateRefresh();
+          const index = this.records.findIndex((item) => item.id === target.id);
+          if (index >= 0 && renamed.updated_at >= this.records[index].updated_at) this.records[index] = renamed;
+          if (this.current?.id === target.id && renamed.updated_at >= this.current.updated_at) this.current = renamed;
+          this.toast('Session renamed.');
+        }
+      } catch (error) {
+        if (error.name !== 'AbortError' && this.renameTarget === target) $('#session-rename-error').textContent = error.message;
+      } finally {
+        this.mutating.delete(target.id);
+        if (this.renameTarget === target) {
+          $('#session-name-input').disabled = false;
+          $('#session-rename-save').disabled = false;
+          $('#session-rename-save').textContent = 'Save name';
+          $('#session-rename-cancel').disabled = false;
+        }
+        this.renderList();
+        this.renderStatus();
+      }
+    }
+
+    async deleteSession(record) {
+      if (!record || this.isBusy(record.id) || document.querySelector('dialog[open]')) return;
+      const live = ['active', 'starting'].includes(record.status);
+      const confirmed = await this.confirmAction(`Delete ${record.name}?`,
+        live ? 'End this shell and permanently delete the session and its saved terminal output.'
+          : 'Permanently delete this session and its saved terminal output.', 'Delete session', true);
+      if (!confirmed) return;
+      const token = this.getToken();
+      this.mutating.set(record.id, 'delete');
+      this.showError('');
+      this.renderList();
+      this.renderStatus();
+      try {
+        await this.api(path(record.vm_id, `/${encodeURIComponent(record.id)}`), { method: 'DELETE' });
+        if (token === this.getToken() && this.vm?.id === record.vm_id) {
+          this.removeRecord(record.vm_id, record.id);
+          this.toast('Session deleted.');
+        }
+      } catch (error) {
+        if (error.name !== 'AbortError' && token === this.getToken() && this.vm?.id === record.vm_id) this.showError(error.message);
+      } finally {
+        this.mutating.delete(record.id);
+        this.renderList();
+        this.renderStatus();
+      }
+    }
+
+    removeRecord(vmId, sessionId) {
+      if (this.vm?.id !== vmId) return;
+      this.invalidateRefresh();
+      this.records = this.records.filter((record) => record.id !== sessionId);
+      $('#vm-session-count').textContent = this.records.length;
+      if (this.renameTarget?.id === sessionId) {
+        $('#session-rename-dialog').close();
+        this.renameTarget = null;
+      }
+      if (this.selectedId === sessionId || this.current?.id === sessionId) {
+        this.detachTerminal();
+        this.selectedId = null;
+        if (this.visible && this.isSelectedVm(vmId)) location.hash = href(vmId);
+      }
+      this.showError('');
+      this.renderList();
+      this.renderStatus();
+    }
+
     detachTerminal() {
       this.generation += 1;
       window.clearTimeout(this.reconnectTimer);
@@ -348,6 +504,8 @@
     }
 
     leave() {
+      if ($('#session-rename-dialog').open) $('#session-rename-dialog').close();
+      this.renameTarget = null;
       this.sequence += 1;
       this.loading = false;
       this.lastRefresh = 0;

nest/dashboard/static/js/tools.js

@@ -0,0 +1,331 @@
+(function () {
+  'use strict';
+
+  const $ = (selector) => document.querySelector(selector);
+  const active = (job) => ['queued', 'starting', 'waiting', 'installing'].includes(job?.status);
+  const retryable = (item) => ['failed', 'interrupted', 'skipped'].includes(item.status);
+  const labels = { queued: 'Queued', starting: 'Starting machine', waiting: 'Waiting for guest', installing: 'Installing',
+    installed: 'Installed', already_installed: 'Already installed', succeeded: 'Complete', failed: 'Failed',
+    interrupted: 'Interrupted', skipped: 'Not attempted' };
+
+  window.NestTools = class NestTools {
+    constructor({ api, esc, toast, formatDate, onSelectionChange, onMutation }) {
+      Object.assign(this, { api, esc, toast, formatDate, onSelectionChange, onMutation });
+      this.catalog = null;
+      this.catalogPromise = null;
+      this.epoch = 0;
+      this.generation = 0;
+      this.createSelected = new Set();
+      this.selected = new Set();
+      this.cache = new WeakMap();
+      this.createBusy = false;
+      this.vm = null;
+      this.data = null;
+      this.timer = null;
+      this.loading = false;
+      this.submitting = false;
+      this.revision = 0;
+      this.probed = false;
+      this.lastProbe = 0;
+      this.viewing = null;
+      this.historyJob = null;
+      this.pendingRequest = null;
+      $('#create-tools-list').addEventListener('change', (event) => {
+        const input = event.target.closest('[data-create-tool]');
+        if (!input) return;
+        if (input.checked) this.createSelected.add(input.dataset.createTool);
+        else this.createSelected.delete(input.dataset.createTool);
+        input.closest('label').classList.toggle('is-selected', input.checked);
+        this.onSelectionChange();
+      });
+      $('#create-tools-refresh').addEventListener('click', () => this.mountCreate());
+      $('#tools-refresh').addEventListener('click', () => this.load(true));
+      $('#tools-install-selected').addEventListener('click', () => this.submit([...this.selected]));
+      $('#tools-retry').addEventListener('click', () => this.submit(this.retryKeys()));
+      $('#tools-restart').addEventListener('click', () => this.submit(this.selected.size ? [...this.selected] : this.retryKeys(), true));
+      $('#tools-job-select').addEventListener('change', (event) => this.viewJob(event.target.value));
+      $('#tools-list').addEventListener('change', (event) => {
+        const input = event.target.closest('[data-tool-select]');
+        if (!input) return;
+        if (input.checked) this.selected.add(input.dataset.toolSelect);
+        else this.selected.delete(input.dataset.toolSelect);
+        this.render();
+      });
+      $('#tools-list').addEventListener('click', (event) => {
+        const button = event.target.closest('[data-install-tool]');
+        if (button && !button.disabled) this.submit([button.dataset.installTool]);
+      });
+    }
+
+    html(selector, value) {
+      const element = $(selector);
+      if (this.cache.get(element) !== value) {
+        element.innerHTML = value;
+        this.cache.set(element, value);
+      }
+    }
+
+    variants() {
+      return (this.catalog || []).flatMap((tool) => (tool.versions.length ? tool.versions : [null]).map((version) => ({
+        ...tool, version, key: version ? `${tool.id}-${version}` : tool.id,
+        title: version ? `${tool.name} ${version}` : tool.name,
+      })));
+    }
+
+    async loadCatalog() {
+      if (this.catalog) return this.catalog;
+      if (this.catalogPromise) return this.catalogPromise;
+      const epoch = this.epoch;
+      const request = this.api('/tools').then((catalog) => {
+        if (!Array.isArray(catalog) || catalog.some((tool) => !tool.id || !Array.isArray(tool.versions))) {
+          throw new Error('The tool catalog response was not recognized.');
+        }
+        if (epoch === this.epoch) this.catalog = catalog;
+        return catalog;
+      }).finally(() => { if (this.catalogPromise === request) this.catalogPromise = null; });
+      this.catalogPromise = request;
+      return request;
+    }
+
+    async mountCreate() {
+      const epoch = this.epoch;
+      $('#create-tools-error').textContent = '';
+      try {
+        await this.loadCatalog();
+        if (epoch !== this.epoch) return;
+        this.html('#create-tools-list', this.variants().map((tool) => `<label class="tool-choice${this.createSelected.has(tool.key) ? ' is-selected' : ''}">
+          <input type="checkbox" data-create-tool="${this.esc(tool.key)}" ${this.createSelected.has(tool.key) ? 'checked' : ''} ${this.createBusy ? 'disabled' : ''} />
+          <span><strong>${this.esc(tool.title)}</strong><small>${this.esc(tool.version ? 'pip + venv' : tool.version_label)}</small></span></label>`).join(''));
+      } catch (error) {
+        if (epoch === this.epoch && error.name !== 'AbortError') $('#create-tools-error').textContent = error.message;
+      }
+    }
+
+    createSelections() {
+      return this.variants().filter((tool) => this.createSelected.has(tool.key)).map(({ id, version }) => ({ id, version }));
+    }
+
+    createSummary() {
+      return this.variants().filter((tool) => this.createSelected.has(tool.key)).map((tool) => tool.title).join(', ') || 'None selected';
+    }
+
+    resetCreate() {
+      this.createSelected.clear();
+      document.querySelectorAll('[data-create-tool]').forEach((input) => {
+        input.checked = false;
+        input.closest('label').classList.remove('is-selected');
+      });
+    }
+
+    setCreateBusy(busy) {
+      this.createBusy = busy;
+      document.querySelectorAll('[data-create-tool]').forEach((input) => { input.disabled = busy; });
+    }
+
+    sync(vm, visible) {
+      if (!visible) return;
+      if (this.vm?.id !== vm.id) {
+        this.leave();
+        this.vm = vm;
+        this.html('#tools-list', '<div class="loading-state">Bringing your tools into view…</div>');
+        $('#tools-job').hidden = true;
+        $('#tools-error').textContent = '';
+        this.load();
+      } else {
+        const started = this.vm.observed_status !== 'running' && vm.observed_status === 'running';
+        this.vm = vm;
+        this.render();
+        if (started && !this.loading) this.load(true);
+      }
+    }
+
+    leave() {
+      this.generation += 1;
+      window.clearTimeout(this.timer);
+      this.timer = null;
+      this.vm = null;
+      this.data = null;
+      this.loading = false;
+      this.submitting = false;
+      this.probed = false;
+      this.selected.clear();
+      this.viewing = null;
+      this.historyJob = null;
+      this.pendingRequest = null;
+    }
+
+    reset() {
+      this.leave();
+      this.epoch += 1;
+      this.catalog = null;
+      this.catalogPromise = null;
+      this.resetCreate();
+    }
+
+    schedule() {
+      window.clearTimeout(this.timer);
+      if (!this.vm) return;
+      this.timer = window.setTimeout(() => {
+        if (document.hidden) { this.schedule(); return; }
+        this.load(Boolean(this.data?.guest_busy && Date.now() - this.lastProbe > 20000));
+      }, active(this.data?.latest_job) ? 1500 : 5000);
+    }
+
+    async load(refresh = false) {
+      if (!this.vm || this.loading) return;
+      const id = this.vm.id;
+      const generation = this.generation;
+      const revision = this.revision;
+      const wasActive = active(this.data?.latest_job);
+      this.loading = true;
+      if (refresh) { this.probed = true; this.lastProbe = Date.now(); }
+      $('#tools-refresh').disabled = true;
+      if (refresh) $('#tools-note').textContent = 'Checking installed tools inside this machine…';
+      let probeNext = false;
+      try {
+        const [, data] = await Promise.all([this.loadCatalog(), this.api(`/vms/${encodeURIComponent(id)}/tools${refresh ? '?refresh=true' : ''}`, { timeout: 100000 })]);
+        if (generation !== this.generation || revision !== this.revision) return;
+        this.data = data;
+        if (!data?.vm_id || data.vm_id !== id) throw new Error('The tool status response was not recognized.');
+        if (this.viewing && !data.jobs.some((job) => job.id === this.viewing)) { this.viewing = null; this.historyJob = null; }
+        this.selected = new Set([...this.selected].filter((key) => !data.inventory.some((item) => item.key === key && item.installed)));
+        $('#tools-error').textContent = data.error || '';
+        probeNext = !active(data.latest_job) && this.vm.observed_status === 'running'
+          && ((wasActive && !refresh) || (!this.probed && data.stale));
+        this.render();
+      } catch (error) {
+        if (generation === this.generation && error.name !== 'AbortError') $('#tools-error').textContent = error.message;
+      } finally {
+        if (generation === this.generation) {
+          this.loading = false;
+          $('#tools-refresh').disabled = false;
+          if (probeNext) this.load(true);
+          else this.schedule();
+        }
+      }
+    }
+
+    retryKeys() {
+      return (this.data?.latest_job?.items || []).filter(retryable).map((item) => item.key);
+    }
+
+    canInstall() {
+      const status = this.vm?.observed_status || this.vm?.status;
+      return ['running', 'stopped'].includes(status) && !this.vm.operation && !this.vm.pending_resize
+        && (status !== 'stopped' || this.vm.ports !== null) && this.data?.supported !== false;
+    }
+
+    render() {
+      if (!this.vm || !this.catalog) return;
+      const job = this.data?.latest_job;
+      const occupied = this.submitting || active(job) || this.data?.guest_busy;
+      const disabled = occupied || !this.canInstall();
+      const stopped = (this.vm.observed_status || this.vm.status) === 'stopped';
+      const focused = document.activeElement?.dataset.toolSelect;
+      const inventory = new Map((this.data?.inventory || []).map((item) => [item.key, item]));
+      const items = new Map((job?.items || []).map((item) => [item.key, item]));
+      this.html('#tools-list', this.variants().map((tool) => {
+        const found = inventory.get(tool.key);
+        const item = items.get(tool.key);
+        const installing = active(job) && item && ['queued', 'installing'].includes(item.status);
+        const installed = found?.installed && !installing;
+        const failed = item && retryable(item) && !installed && !installing;
+        const state = installing ? labels[item.status] : installed ? `Installed · ${found.detected_version}` : failed ? labels[item.status]
+          : found?.detected_version ? `Found ${found.detected_version}` : found ? 'Available' : 'Not checked yet';
+        const caption = stopped ? 'Start & install' : failed ? 'Retry' : 'Install';
+        return `<article class="tool-card${installed ? ' tool-installed' : ''}">
+          <label class="tool-card-heading"><input type="checkbox" data-tool-select="${this.esc(tool.key)}" aria-label="Select ${this.esc(tool.title)}" ${this.selected.has(tool.key) ? 'checked' : ''} ${disabled || installed ? 'disabled' : ''} />
+          <span><strong>${this.esc(tool.title)}</strong><small>${this.esc(tool.version ? `Python ${tool.version} · pip + venv` : tool.version_label)}</small></span></label>
+          <p>${this.esc(tool.description)}</p><small class="tool-source">${this.esc(tool.source)}</small>
+          <div class="tool-card-footer"><span class="tool-state ${installed ? 'tool-state-good' : failed ? 'tool-state-error' : installing ? 'tool-state-active' : ''}">${this.esc(state)}</span>
+          <button class="button button-small" type="button" data-install-tool="${this.esc(tool.key)}" ${disabled || installed ? 'disabled' : ''}>${installed ? 'Installed' : this.esc(caption)}</button></div>
+          ${installed && found.usage ? `<details class="tool-usage"><summary>Get started</summary><code>${this.esc(found.usage)}</code><small>Run these commands in this machine’s terminal.</small></details>` : ''}</article>`;
+      }).join(''));
+      if (focused) document.querySelector(`[data-tool-select="${CSS.escape(focused)}"]`)?.focus({ preventScroll: true });
+      $('#tools-selected-count').textContent = `${this.selected.size} selected`;
+      $('#tools-install-selected').textContent = this.submitting ? 'Queuing…' : stopped ? 'Start & install selected' : 'Install selected';
+      $('#tools-install-selected').disabled = disabled || !this.selected.size;
+      $('#tools-retry').hidden = !this.retryKeys().length || active(job);
+      $('#tools-retry').disabled = disabled;
+      $('#tools-restart').hidden = stopped || !this.data?.error || this.data.agent_ready || active(job);
+      $('#tools-restart').disabled = disabled || (!this.selected.size && !this.retryKeys().length);
+      $('#tools-note').textContent = this.data?.guest_busy
+        ? 'An installation is still running inside the guest. Its status will be checked again shortly.'
+        : stopped ? 'Start & install boots this machine and installs your selected tools. Versions shown are last observed.'
+          : this.data?.checked_at ? `Last checked ${this.formatDate(this.data.checked_at)}${this.data.stale ? ' · cached observations' : ''}. Use Get started for commands and shell initialization.`
+            : 'Tools are installed directly inside this VM. Guest internet access is required for downloads.';
+      this.renderJob();
+    }
+
+    renderJob() {
+      const job = this.historyJob || this.data?.latest_job;
+      $('#tools-job').hidden = !job;
+      if (!job) return;
+      $('#tools-job-title').textContent = labels[job.status] || job.status;
+      $('#tools-job-message').textContent = job.error || (active(job) ? 'You can leave this page. Installation continues in the background.' : 'Tool installation results are saved with this machine.');
+      const complete = job.items.filter((item) => !['queued', 'installing'].includes(item.status)).length;
+      $('#tools-progress').max = job.items.length;
+      $('#tools-progress').value = complete;
+      $('#tools-progress').setAttribute('aria-label', `${complete} of ${job.items.length} tools processed`);
+      const names = new Map(this.variants().map((tool) => [tool.key, tool.title]));
+      const open = new Set([...document.querySelectorAll('#tools-job-items details[open]')].map((item) => item.dataset.output));
+      this.html('#tools-job-items', job.items.map((item) => `<li><details data-output="${this.esc(item.key)}" ${open.has(item.key) ? 'open' : ''}>
+        <summary><strong>${this.esc(names.get(item.key) || item.key)}</strong><span class="tool-state ${retryable(item) ? 'tool-state-error' : ''}">${this.esc(labels[item.status] || item.status)}</span></summary>
+        ${item.error ? `<p class="form-error">${this.esc(item.error)}</p>` : ''}<pre class="tool-output" tabindex="0">${this.esc(item.output || 'No output recorded.')}</pre>
+        ${item.truncated ? '<p class="panel-footnote">Showing the bounded tail of installation output.</p>' : ''}</details></li>`).join(''));
+      this.html('#tools-job-select', (this.data.jobs || []).map((item) => `<option value="${this.esc(item.id)}">${this.esc(this.formatDate(item.created_at))} · ${this.esc(labels[item.status] || item.status)}</option>`).join(''));
+      $('#tools-job-select').value = job.id;
+    }
+
+    async viewJob(id) {
+      if (!this.vm) return;
+      const generation = this.generation;
+      this.viewing = id === this.data?.latest_job?.id ? null : id;
+      this.historyJob = null;
+      if (!this.viewing) { this.renderJob(); return; }
+      try {
+        const job = await this.api(`/vms/${encodeURIComponent(this.vm.id)}/tools/jobs/${encodeURIComponent(id)}`);
+        if (generation !== this.generation || this.viewing !== id) return;
+        this.historyJob = job;
+        this.renderJob();
+      } catch (error) {
+        if (generation === this.generation && error.name !== 'AbortError') this.toast(error.message, true);
+      }
+    }
+
+    async submit(keys, restart = false) {
+      if (!this.vm || this.submitting || active(this.data?.latest_job) || !this.canInstall()) return;
+      const tools = this.variants().filter((tool) => keys.includes(tool.key)).map(({ id, version }) => ({ id, version }));
+      if (!tools.length) return;
+      const id = this.vm.id;
+      const generation = this.generation;
+      const payload = { tools, start_if_stopped: true, restart };
+      const signature = JSON.stringify(payload);
+      if (this.pendingRequest?.signature !== signature) this.pendingRequest = { signature, id: crypto.randomUUID() };
+      payload.request_id = this.pendingRequest.id;
+      this.submitting = true;
+      this.revision += 1;
+      $('#tools-error').textContent = '';
+      this.render();
+      try {
+        const job = await this.api(`/vms/${encodeURIComponent(id)}/tools/install`, { method: 'POST', body: payload });
+        if (generation !== this.generation) return;
+        this.pendingRequest = null;
+        this.selected.clear();
+        this.viewing = null;
+        this.historyJob = null;
+        if (this.data) { this.data.latest_job = job; this.data.jobs = [{ id: job.id, status: job.status, created_at: job.created_at }, ...this.data.jobs.filter((item) => item.id !== job.id)]; }
+        this.toast('Tool installation queued. You can follow its progress here.');
+        this.onMutation(id);
+      } catch (error) {
+        if (generation === this.generation && error.name !== 'AbortError') $('#tools-error').textContent = error.message;
+      } finally {
+        if (generation === this.generation) {
+          this.submitting = false;
+          this.render();
+          this.load();
+        }
+      }
+    }
+  };
+})();

nest/session_routes.py

@@ -10,15 +10,16 @@ from contextlib import suppress
 from typing import Literal
 from urllib.parse import urlsplit
 
-from fastapi import APIRouter, Depends, Request, WebSocket, WebSocketDisconnect
-from pydantic import BaseModel, ConfigDict, Field, StrictInt
+from fastapi import APIRouter, Depends, Request, Response, WebSocket, WebSocketDisconnect
+from pydantic import BaseModel, ConfigDict, Field, StrictInt, StrictStr
 
 from uvm.app import Application
-from uvm.errors import UvmError
+from uvm.errors import NotFoundError, UvmError
 from uvm.guest_agent import MAX_TERMINAL_FRAME, ProtocolError, terminal_data
 from uvm.routers.dependencies import get_authorized_application
 
 from .sessions import SessionManager
+from .session_store import MAX_SESSION_NAME_LENGTH
 
 
 router = APIRouter(prefix="/vms/{vm_id}/sessions", tags=["sessions"])
@@ -48,6 +49,11 @@ class SessionResponse(BaseModel):
     truncated: bool
 
 
+class SessionRename(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+    name: StrictStr = Field(min_length=1, max_length=MAX_SESSION_NAME_LENGTH)
+
+
 def manager(request: Request, _application: Application = Depends(get_authorized_application)) -> SessionManager:
     return request.app.state.nest_sessions
 
@@ -72,6 +78,17 @@ async def close_session(vm_id: str, session_id: str, sessions: SessionManager =
     return await sessions.end(vm_id, session_id)
 
 
+@router.patch("/{session_id}", response_model=SessionResponse)
+async def rename_session(vm_id: str, session_id: str, body: SessionRename, sessions: SessionManager = Depends(manager)):
+    return await sessions.rename(vm_id, session_id, body.name)
+
+
+@router.delete("/{session_id}", status_code=204)
+async def delete_session(vm_id: str, session_id: str, sessions: SessionManager = Depends(manager)) -> Response:
+    await sessions.delete(vm_id, session_id)
+    return Response(status_code=204)
+
+
 def _same_origin(websocket: WebSocket) -> bool:
     origin = websocket.headers.get("origin")
     if origin is None:
@@ -101,6 +118,7 @@ async def terminal(websocket: WebSocket, vm_id: str, session_id: str):
         return
     await websocket.accept()
     tasks = []
+    attached = False
     try:
         # Browser WebSockets cannot set X-UVM-Token. Authenticate in the first
         # frame so tokens never appear in URLs, access logs, or session history.
@@ -113,6 +131,7 @@ async def terminal(websocket: WebSocket, vm_id: str, session_id: str):
             return
         sessions = websocket.app.state.nest_sessions
         await sessions.get(vm_id, session_id)
+        attached = True
         offset = auth.get("offset", 0)
         acknowledgements = asyncio.Queue(maxsize=1)
 
@@ -133,7 +152,7 @@ async def terminal(websocket: WebSocket, vm_id: str, session_id: str):
                     if type(acknowledged) is not int or acknowledged != cursor:
                         raise ProtocolError("invalid terminal output acknowledgement")
                     continue
-                if record["status"] not in {"active", "starting"}:
+                if record["status"] not in {"active", "starting"} and not sessions.is_deleting(session_id):
                     return
                 await sessions.wait(session_id)
 
@@ -157,6 +176,12 @@ async def terminal(websocket: WebSocket, vm_id: str, session_id: str):
             task.result()
     except WebSocketDisconnect:
         pass
+    except NotFoundError as error:
+        with suppress(WebSocketDisconnect, RuntimeError):
+            await websocket.send_json(
+                {"type": "deleted", "session_id": session_id} if attached
+                else {"type": "error", "message": str(error)}
+            )
     except (ValueError, KeyError, TypeError, UnicodeError, RecursionError, TimeoutError):
         with suppress(WebSocketDisconnect, RuntimeError):
             await websocket.send_json({"type": "error", "message": "Invalid terminal message or authentication timeout."})

nest/session_store.py

@@ -16,6 +16,16 @@ from uvm.errors import ConflictError, NotFoundError, StateError, ValidationError
 
 
 MAX_TRANSCRIPT_BYTES = 8 * 1024 * 1024
+MAX_SESSION_NAME_LENGTH = 80
+
+
+def validate_session_name(value: object) -> str:
+    if not isinstance(value, str):
+        raise ValidationError("Session name must be text.")
+    name = value.strip()
+    if not 1 <= len(name) <= MAX_SESSION_NAME_LENGTH or not name.isprintable():
+        raise ValidationError(f"Session name must contain 1–{MAX_SESSION_NAME_LENGTH} printable characters.")
+    return name
 
 
 class SessionStore:
@@ -100,7 +110,10 @@ class SessionStore:
         session_id = "session-" + uuid.uuid4().hex
         now = time.time()
         with self._db() as db:
-            number = db.execute("SELECT COUNT(*) FROM sessions WHERE vm_id=?", (vm_id,)).fetchone()[0] + 1
+            names = [row[0] for row in db.execute("SELECT name FROM sessions WHERE vm_id=?", (vm_id,))]
+            number = len(names) + 1
+            while f"Session {number}" in names:
+                number += 1
             db.execute("""INSERT INTO sessions
                 (id, vm_id, name, username, status, created_at, updated_at, rows, cols)
                 VALUES (?, ?, ?, '', 'starting', ?, ?, ?, ?)""",
@@ -117,6 +130,24 @@ class SessionStore:
         with self._db() as db:
             return self._record(db.execute("SELECT * FROM sessions WHERE vm_id=? AND id=?", (vm_id, session_id)).fetchone())
 
+    def rename(self, vm_id: str, session_id: str, name: str) -> dict:
+        name = validate_session_name(name)
+        with self._db() as db:
+            self._record(db.execute("SELECT * FROM sessions WHERE vm_id=? AND id=?", (vm_id, session_id)).fetchone())
+            db.execute("UPDATE sessions SET name=?, updated_at=? WHERE vm_id=? AND id=?",
+                       (name, time.time(), vm_id, session_id))
+            return self._record(db.execute("SELECT * FROM sessions WHERE vm_id=? AND id=?", (vm_id, session_id)).fetchone())
+
+    def delete(self, vm_id: str, session_id: str) -> None:
+        with self._db() as db:
+            record = self._record(db.execute("SELECT * FROM sessions WHERE vm_id=? AND id=?", (vm_id, session_id)).fetchone())
+            if record["status"] in {"starting", "active"}:
+                raise ConflictError("End the session before deleting its history.")
+            # Both tables are removed in one transaction, after the output reader
+            # has finished. Explicit cleanup also supports existing databases.
+            db.execute("DELETE FROM output WHERE session_id=?", (session_id,))
+            db.execute("DELETE FROM sessions WHERE vm_id=? AND id=?", (vm_id, session_id))
+
     def activate(self, session_id: str, username: str) -> None:
         with self._db() as db:
             db.execute("UPDATE sessions SET status='active', username=?, updated_at=? WHERE id=?",

nest/sessions.py

@@ -10,7 +10,7 @@ from uvm.app import Application
 from uvm.errors import ConflictError, GuestExecutionError, NotFoundError, StateError, UvmError
 from uvm.guest_terminal import GuestTerminal, validate_size
 
-from .session_store import SessionStore
+from .session_store import SessionStore, validate_session_name
 
 
 async def _persist(function, *arguments):
@@ -38,7 +38,8 @@ class SessionManager:
         self._store: SessionStore | None = None
         self._live: dict[str, _LiveSession] = {}
         self._pending: set[asyncio.Task] = set()
-        self._create_lock = asyncio.Lock()
+        self._deleting: set[str] = set()
+        self._session_lock = asyncio.Lock()
         self._closing = False
 
     def _history(self) -> SessionStore:
@@ -68,7 +69,11 @@ class SessionManager:
         await self._vm(vm_id)
         history = self._history()
         # A dropped HTTP request must not abandon a just-created guest connection.
-        task = asyncio.create_task(self._create(history, vm_id, rows, cols))
+        return await self._run_pending(self._create(history, vm_id, rows, cols))
+
+    async def _run_pending(self, operation):
+        """Finish accepted mutations even if their HTTP request disconnects."""
+        task = asyncio.create_task(operation)
         self._pending.add(task)
 
         def completed(future):
@@ -80,7 +85,7 @@ class SessionManager:
         return await asyncio.shield(task)
 
     async def _create(self, history: SessionStore, vm_id: str, rows: int, cols: int) -> dict:
-        async with self._create_lock:
+        async with self._session_lock:
             if self._closing:
                 raise ConflictError("Nest is shutting down.")
             if len(self._live) >= 32:
@@ -165,7 +170,15 @@ class SessionManager:
             raise GuestExecutionError("The terminal connection was interrupted.", outcome_unknown=True) from error
 
     async def end(self, vm_id: str, session_id: str) -> dict:
-        record = await self.get(vm_id, session_id)
+        await self.get(vm_id, session_id)
+        return await self._run_pending(self._end(self._history(), vm_id, session_id))
+
+    async def _end(self, history: SessionStore, vm_id: str, session_id: str) -> dict:
+        async with self._session_lock:
+            return await self._end_locked(history, vm_id, session_id)
+
+    async def _end_locked(self, history: SessionStore, vm_id: str, session_id: str) -> dict:
+        record = await asyncio.to_thread(history.get, vm_id, session_id)
         live = self._live.get(session_id)
         if live is not None:
             try:
@@ -174,9 +187,42 @@ class SessionManager:
             except (OSError, UvmError):
                 live.task.cancel()
                 await asyncio.gather(live.task, return_exceptions=True)
-            record = await self.get(vm_id, session_id)
+        elif record["status"] in {"starting", "active"}:
+            # The session lock excludes creation here. A missing channel can
+            # only be an interrupted session whose final metadata write failed.
+            await _persist(history.finish, session_id, "interrupted", None, "This shell is no longer connected.")
+        return await asyncio.to_thread(history.get, vm_id, session_id)
+
+    async def rename(self, vm_id: str, session_id: str, name: str) -> dict:
+        name = validate_session_name(name)
+        await self.get(vm_id, session_id)
+        return await self._run_pending(self._rename(self._history(), vm_id, session_id, name))
+
+    async def _rename(self, history: SessionStore, vm_id: str, session_id: str, name: str) -> dict:
+        record = await _persist(history.rename, vm_id, session_id, name)
+        live = self._live.get(session_id)
+        if live is not None:
+            live.changed.set()
         return record
 
+    async def delete(self, vm_id: str, session_id: str) -> None:
+        await self.get(vm_id, session_id)
+        await self._run_pending(self._delete(self._history(), vm_id, session_id))
+
+    async def _delete(self, history: SessionStore, vm_id: str, session_id: str) -> None:
+        # Creation must finish before deletion can close its channel. Join the
+        # output reader (including pending disk writes) before removing history.
+        async with self._session_lock:
+            self._deleting.add(session_id)
+            try:
+                await self._end_locked(history, vm_id, session_id)
+                await _persist(history.delete, vm_id, session_id)
+            finally:
+                self._deleting.discard(session_id)
+
+    def is_deleting(self, session_id: str) -> bool:
+        return session_id in self._deleting
+
     async def shutdown(self) -> None:
         self._closing = True
         await asyncio.gather(*self._pending, return_exceptions=True)

tests/browser_sessions.cjs

@@ -30,7 +30,8 @@ const { chromium } = require('playwright');
     }
     assert.ok(ready, `Fixture server did not start: ${logs}`);
     browser = await chromium.launch({ headless: true });
-    page = await browser.newPage({ viewport: { width: 1440, height: 1100 } });
+    const context = await browser.newContext({ viewport: { width: 1440, height: 1100 } });
+    page = await context.newPage();
     const errors = [];
     page.on('pageerror', (error) => errors.push(error.message));
     const connected = () => page.waitForFunction(() => document.querySelector('#session-status').textContent === 'Connected');
@@ -51,15 +52,49 @@ const { chromium } = require('playwright');
     await output('result:echo alpha');
     assert.equal(await page.locator('.session-item').count(), 1);
 
+    // Rename is cancellable, validated, persisted, and preserves the live widget.
+    const firstId = new URL(firstUrl).hash.split('/').at(-1);
+    const firstRow = () => page.locator(`.session-item[data-session-id="${firstId}"]`);
+    await page.locator('#session-rename').click();
+    await page.locator('#session-name-input').fill('Discarded name');
+    await page.locator('#session-rename-cancel').click();
+    assert.equal(await firstRow().locator('strong').textContent(), 'Session 1');
+    await page.evaluate(() => { window.originalTerminal = document.querySelector('#session-terminal .xterm'); });
+    await page.locator('#session-rename').click();
+    await page.locator('#session-name-input').fill('   ');
+    await page.locator('#session-rename-save').click();
+    assert.ok(await page.locator('#session-rename-dialog').isVisible());
+    assert.match(await page.locator('#session-rename-error').textContent(), /Enter a session name/);
+    await page.locator('#session-name-input').fill('  Deployment  ');
+    await page.locator('#session-rename-save').click();
+    await page.locator('#session-rename-dialog').waitFor({ state: 'hidden' });
+    await page.waitForFunction(() => document.querySelector('#session-title').textContent.startsWith('Deployment'));
+    assert.equal(page.url(), firstUrl);
+    assert.ok(await page.evaluate(() => window.originalTerminal === document.querySelector('#session-terminal .xterm')));
+    await page.reload();
+    await connected();
+    await output('result:echo alpha');
+    assert.equal(await firstRow().locator('strong').textContent(), 'Deployment');
+
     await page.locator('#detail-actions [data-action="launch-session"]').click();
     await page.waitForURL((url) => url.href !== firstUrl);
     await connected();
+    const secondUrl = page.url();
     assert.equal(await page.locator('.session-item').count(), 2);
     assert.ok(!(await page.locator('#session-terminal').textContent()).includes('result:echo alpha'));
     await page.keyboard.type('echo beta');
     await page.keyboard.press('Enter');
     await output('result:echo beta');
-    await page.locator('.session-item').filter({ hasText: 'Session 1' }).click();
+    // History actions also work without selecting the affected session.
+    await firstRow().locator('[data-session-action="rename"]').click();
+    await page.locator('#session-name-input').fill('Operations <alpha>');
+    await page.locator('#session-rename-save').click();
+    await page.locator('#session-rename-dialog').waitFor({ state: 'hidden' });
+    assert.equal(page.url(), secondUrl);
+    assert.equal(await firstRow().locator('strong').textContent(), 'Operations <alpha>');
+    assert.equal(await firstRow().locator('alpha').count(), 0);
+    await output('result:echo beta');
+    await firstRow().locator('.session-link').click();
     await connected();
     await output('result:echo alpha');
     assert.ok(!(await page.locator('#session-terminal').textContent()).includes('result:echo beta'));
@@ -80,14 +115,59 @@ const { chromium } = require('playwright');
     await page.reload();
     await output('result:echo alpha');
     assert.ok(await page.locator('#session-end').isDisabled());
+    await page.locator('#session-rename').click();
+    await page.locator('#session-name-input').fill('Archived run');
+    await page.locator('#session-rename-save').click();
+    await page.locator('#session-rename-dialog').waitFor({ state: 'hidden' });
+    await page.reload();
+    await output('result:echo alpha');
+    assert.equal(await firstRow().locator('strong').textContent(), 'Archived run');
 
     await page.setViewportSize({ width: 390, height: 844 });
     await page.waitForTimeout(300);
     const fits = await page.evaluate(() => document.documentElement.scrollWidth <= window.innerWidth);
     assert.ok(fits, 'mobile layout overflows the viewport');
-    assert.deepEqual(errors, []);
     if (process.env.NEST_BROWSER_SCREENSHOT) await page.screenshot({ path: process.env.NEST_BROWSER_SCREENSHOT, fullPage: true });
-    console.log('Browser sessions passed: launch, terminal input, reload/reconnect, multiple sessions, VM isolation, closed history, and mobile layout.');
+
+    // Cancel preserves history; confirmation removes an ended session and clears selection.
+    await page.locator('#session-delete').click();
+    await page.locator('#confirm-dialog button[value="cancel"]').click();
+    assert.equal(await page.locator('.session-item').count(), 2);
+    await firstRow().locator('[data-session-action="delete"]').click();
+    await page.locator('#confirm-ok').click();
+    await page.waitForURL((url) => url.hash === '#vm/vm-test/sessions');
+    await page.waitForFunction(() => document.querySelector('#vm-session-count').textContent === '1');
+    assert.ok(await page.locator('#session-terminal').isHidden());
+    await page.reload();
+    await page.waitForFunction(() => document.querySelector('#vm-session-count').textContent === '1');
+    assert.equal(await firstRow().count(), 0);
+
+    // Deleting an active session clears other attached browser views as well.
+    await page.goto(secondUrl);
+    await connected();
+    await output('result:echo beta');
+    const observer = await page.context().newPage();
+    observer.on('pageerror', (error) => errors.push(error.message));
+    await observer.goto(secondUrl);
+    await observer.locator('#token-input').fill('browser-test-token');
+    await observer.locator('#auth-btn').click();
+    await observer.waitForFunction(() => document.querySelector('#session-status').textContent === 'Connected');
+    await page.locator('#session-rename').click();
+    await page.locator('#session-name-input').fill('Second workspace');
+    await page.locator('#session-rename-save').click();
+    await page.locator('#session-rename-dialog').waitFor({ state: 'hidden' });
+    await observer.waitForFunction(() => document.querySelector('#session-title').textContent.startsWith('Second workspace'));
+    await page.locator('#session-delete').click();
+    await page.locator('#confirm-ok').click();
+    await page.waitForFunction(() => document.querySelector('#vm-session-count').textContent === '0');
+    await observer.waitForURL((url) => url.hash === '#vm/vm-test/sessions');
+    assert.ok(await observer.locator('#session-terminal').isHidden());
+    await page.reload();
+    await page.waitForFunction(() => document.querySelector('#vm-session-count').textContent === '0');
+    assert.equal(await page.locator('.session-item').count(), 0);
+    assert.deepEqual(errors, []);
+    await observer.close();
+    console.log('Browser sessions passed: terminal/reconnect, VM isolation, persistent rename, active/ended deletion, cancellation, live browser updates, and mobile layout.');
   } catch (error) {
     if (page) console.error('Browser state:', await page.locator('#sessions-error').textContent(), await page.locator('#session-status').textContent());
     console.error(logs);

tests/browser_tool_server.py

@@ -0,0 +1,130 @@
+"""Browser fixture with the real API/job workers and simulated guest execution."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import shlex
+import threading
+from dataclasses import replace
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import patch
+
+import uvicorn
+
+from nest.server import create_app
+from tests.browser_session_server import BrowserLifecycle
+from uvm.config import Settings
+from uvm.domain import new_vm_id
+from uvm.errors import ConflictError, GuestExecutionError
+from uvm.guest_exec import ExecResult
+from uvm.state import StateStore
+from uvm.tool_store import ToolJobStore
+from uvm.tools import ToolService
+
+
+class FixtureConnection:
+    def __init__(self):
+        self.closed = threading.Event()
+
+    def close(self):
+        self.closed.set()
+
+
+class ToolBrowserLifecycle(BrowserLifecycle):
+    def __init__(self, settings, state, *, delay=1.2):
+        super().__init__()
+        self.settings, self.state, self.delay = settings, state, delay
+        self.tools = ToolJobStore(settings, state)
+        self.installed = {}
+        self.attempts = {}
+        for index, vm in enumerate(self.records.values()):
+            vm.runtime_owned = True
+            vm.pid, vm.process_start_time, vm.boot_id = index + 100, "1", "fixture-boot"
+            settings.vm_dir(vm.id).mkdir(parents=True, mode=0o700, exist_ok=True)
+        with state.transaction() as document:
+            document.vms.update(self.records)
+
+    def create(self, spec):
+        vm = replace(self.vm, id=new_vm_id(), name=spec.name, username=spec.username, cpu=spec.cpu,
+                     ram_mib=spec.ram_mib, storage_mib=spec.storage_mib, ports=spec.ports)
+        with self.state.operation_lock():
+            self.settings.vm_dir(vm.id).mkdir(mode=0o700)
+            self.records[vm.id] = vm
+            with self.state.transaction() as document:
+                document.vms[vm.id] = vm
+            if spec.tools:
+                self.tools.enqueue_locked(vm.id, spec.tools, source="create")
+        return vm
+
+    def queue_tools(self, vm_id, selections, **options):
+        self.state.get(vm_id)
+        with self.state.operation_lock():
+            return self.tools.enqueue_locked(vm_id, selections, **options)
+
+    def prepare_tool_install(self, vm_id, *, start_if_stopped=False, restart=False):
+        with self.state.operation_lock():
+            vm = self.records[vm_id]
+            if vm.status == "stopped" and not (start_if_stopped or restart):
+                raise ConflictError("start the VM first")
+            vm.status = "running"
+            with self.state.transaction() as document:
+                document.vms[vm.id] = vm
+            return vm
+
+    def execute(self, vm_id, command, *, expected_identity=None, on_connect=None, **_options):
+        self.state.get(vm_id)
+        payload = json.loads(shlex.split(command)[-1])
+        connection = FixtureConnection()
+        if on_connect:
+            on_connect(connection)
+        installed = self.installed.setdefault(vm_id, {"curl"})
+
+        def observation(tool):
+            versions = {"nodejs": "24.21.0", "jdk": "21.0.12", "php": "8.5.11", "conda": "26.5.3", "micromamba": "2.9.0"}
+            version = tool["version"] + ".15" if tool["id"] == "python" else versions.get(tool["id"], "2.0.0")
+            return {"key": tool["key"], "id": tool["id"], "version": tool["version"], "installed": tool["key"] in installed,
+                    "detected_version": version if tool["key"] in installed else None, "error": None, "usage": tool["usage"]}
+
+        if payload["action"] == "probe":
+            result = {"ok": True, "busy": False, "boot_id": "guest-fixture", "supported": True,
+                      "tools": [observation(tool) for tool in payload["tools"]]}
+        else:
+            if connection.closed.wait(self.delay):
+                raise GuestExecutionError("fixture connection closed", outcome_unknown=True)
+            tool = payload["tools"][0]
+            key = (vm_id, tool["key"])
+            self.attempts[key] = self.attempts.get(key, 0) + 1
+            if tool["id"] == "php" and self.attempts[key] == 1:
+                result = {"ok": False, "error": "Simulated mirror failure", "output": "<script>bad()</script>\nTry again later."}
+            else:
+                installed.add(tool["key"])
+                result = {"ok": True, "status": "installed", "tool": observation(tool),
+                          "output": f"Verified and installed {tool['key']} inside the guest."}
+        return ExecResult(0, json.dumps(result), "", False, False)
+
+
+def fixture_application(base, *, delay=1.2):
+    settings = replace(Settings(), base=base, api_token="browser-test-token", tool_workers=2)
+    state = StateStore(settings)
+    lifecycle = ToolBrowserLifecycle(settings, state, delay=delay)
+    return SimpleNamespace(settings=settings, state_store=state, lifecycle=lifecycle,
+                           tools=ToolService(settings, state, lifecycle))
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--base", type=Path, required=True)
+    parser.add_argument("--port", type=int, required=True)
+    args = parser.parse_args()
+    application = fixture_application(args.base)
+    with patch("uvm.tools.require_root"):
+        api = create_app(application, host="127.0.0.1")
+        server = uvicorn.Server(uvicorn.Config(api, host="127.0.0.1", port=args.port, log_level="warning"))
+        api.state.uvicorn_server = server
+        server.run()
+
+
+if __name__ == "__main__":
+    main()

tests/browser_tools.cjs

@@ -0,0 +1,106 @@
+// Optional browser integration: real API/job persistence, simulated guest commands.
+const assert = require('node:assert/strict');
+const { spawn } = require('node:child_process');
+const { mkdtemp, rm } = require('node:fs/promises');
+const os = require('node:os');
+const path = require('node:path');
+const { chromium } = require('playwright');
+
+(async () => {
+  const temporary = await mkdtemp(path.join(os.tmpdir(), 'nest-tool-browser-'));
+  const port = process.env.NEST_BROWSER_TEST_PORT || '18766';
+  const origin = `http://127.0.0.1:${port}`;
+  const server = spawn(process.env.NEST_TEST_PYTHON || 'python3',
+    ['-B', '-m', 'tests.browser_tool_server', '--base', temporary, '--port', port],
+    { cwd: path.resolve(__dirname, '..'), stdio: ['ignore', 'pipe', 'pipe'] });
+  let logs = '';
+  server.stdout.on('data', (data) => { logs = (logs + data).slice(-16000); });
+  server.stderr.on('data', (data) => { logs = (logs + data).slice(-16000); });
+  let browser;
+  let page;
+  try {
+    let ready = false;
+    for (let attempt = 0; attempt < 100; attempt++) {
+      if (server.exitCode !== null) throw new Error(`Fixture server exited: ${logs}`);
+      try { if ((await fetch(`${origin}/health`)).ok) { ready = true; break; } } catch {}
+      await new Promise((resolve) => setTimeout(resolve, 100));
+    }
+    assert.ok(ready, logs);
+    browser = await chromium.launch({ headless: true });
+    page = await browser.newPage({ viewport: { width: 1440, height: 1100 } });
+    const errors = [];
+    page.on('pageerror', (error) => errors.push(error.message));
+    const installed = (key) => page.waitForFunction((value) =>
+      document.querySelector(`[data-install-tool="${value}"]`)?.textContent === 'Installed', key);
+    await page.goto(`${origin}/dashboard#create`);
+    await page.locator('#token-input').fill('browser-test-token');
+    await page.locator('#auth-btn').click();
+    await page.locator('[data-create-tool="python-3.12"]').check();
+    await page.locator('[data-create-tool="git"]').check();
+    await page.locator('[data-create-tool="nodejs"]').check();
+    assert.equal(await page.locator('[data-create-tool]').count(), 12);
+    assert.match(await page.locator('#create-summary-details').textContent(), /Git, Node.js, Python 3.12/);
+    const layout = await page.evaluate(() => ({
+      parent: document.querySelector('.create-tools-section').parentElement.className,
+      children: document.querySelector('#create-form').children.length,
+    }));
+    assert.equal(layout.parent, 'create-fields');
+    assert.equal(layout.children, 2);
+    await page.locator('#create-name').fill('tool-workspace');
+    await page.locator('#create-submit').click();
+    await page.waitForURL((url) => /#vm\/vm-[a-f0-9]+\/tools$/.test(url.hash));
+    const createdUrl = page.url();
+    await page.locator('#tools-job').waitFor({ state: 'visible' });
+    // A reload can occur while the server-owned worker is still installing.
+    await page.reload();
+    await page.waitForFunction(() => document.querySelector('#tools-job-title').textContent === 'Complete');
+    await installed('curl');
+    assert.equal(await page.locator('.tool-installed').count(), 4); // Includes image-provided curl.
+    assert.ok(await page.locator('[data-install-tool="python-3.12"]').isDisabled());
+    await page.locator('#tools-job-items details').first().locator('summary').click();
+    assert.match(await page.locator('#tools-job-items').textContent(), /Verified and installed git/);
+    await page.locator('[data-install-tool="php"]').click();
+    await page.waitForFunction(() => document.querySelector('#tools-job-title').textContent === 'Failed');
+    await page.locator('#tools-job-items details[data-output="php"] summary').click();
+    assert.match(await page.locator('.tool-output').textContent(), /<script>bad/);
+    assert.equal(await page.locator('#tools-job-items script').count(), 0);
+    await page.locator('#tools-retry').click();
+    await page.waitForFunction(() => document.querySelector('#tools-job-title').textContent === 'Complete');
+    await installed('php');
+    assert.ok(await page.locator('[data-install-tool="php"]').isDisabled());
+    assert.equal(await page.locator('#tools-job-select option').count(), 3);
+
+    // The same click boots a stopped VM before installing its selected tool.
+    await page.goto(`${origin}/dashboard#vm/vm-other/tools`);
+    await page.locator('[data-install-tool="htop"]').waitFor();
+    assert.equal(await page.locator('[data-install-tool="htop"]').textContent(), 'Start & install');
+    await page.locator('[data-install-tool="htop"]').click();
+    await installed('htop');
+    await page.waitForFunction(() => document.querySelector('#tools-job-title').textContent === 'Complete');
+    assert.ok(await page.locator('[data-install-tool="htop"]').isDisabled());
+    assert.ok(!(await page.locator('#tools-job-items').textContent()).includes('python-3.12'));
+
+    await page.goto(createdUrl);
+    await page.locator('[data-tool-select="wget"]').check();
+    await page.locator('[data-tool-select="micromamba"]').check();
+    await page.locator('#tools-install-selected').click();
+    await installed('wget');
+    await installed('micromamba');
+    await page.waitForFunction(() => document.querySelector('#tools-job-title').textContent === 'Complete');
+    await page.setViewportSize({ width: 390, height: 844 });
+    await page.waitForTimeout(250);
+    assert.ok(await page.evaluate(() => document.documentElement.scrollWidth <= window.innerWidth), 'mobile layout overflows');
+    if (process.env.NEST_BROWSER_SCREENSHOT) await page.screenshot({ path: process.env.NEST_BROWSER_SCREENSHOT, fullPage: true });
+    assert.deepEqual(errors, []);
+    console.log('Browser tools passed: creation, background/reload, detection, failure/retry, safe output, history, Start & install, VM isolation, multi-select, and mobile layout.');
+  } catch (error) {
+    if (page) console.error('Tool UI:', await page.locator('#tools-error').textContent(), await page.locator('#tools-job-message').textContent());
+    console.error(logs);
+    throw error;
+  } finally {
+    await browser?.close();
+    server.kill('SIGTERM');
+    if (server.exitCode === null) await new Promise((resolve) => server.once('exit', resolve));
+    await rm(temporary, { recursive: true, force: true });
+  }
+})().catch((error) => { console.error(error); process.exitCode = 1; });

tests/integration/test_guest_tools.py

@@ -0,0 +1,96 @@
+"""Opt-in tool installation smoke test against a prepared, disposable microVM."""
+
+from __future__ import annotations
+
+import json
+import os
+import time
+import unittest
+import uuid
+from urllib.parse import quote
+from urllib.request import Request, urlopen
+
+
+@unittest.skipUnless(os.environ.get("UVM_TOOLS_INTEGRATION") == "1",
+                     "set UVM_TOOLS_INTEGRATION=1 on an isolated disposable VM host")
+class GuestToolsIntegrationTests(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.api_url = os.environ.get("UVM_API_URL", "http://127.0.0.1:8000").rstrip("/")
+        cls.token = os.environ["UVM_API_TOKEN"]
+        cls.vm_id = os.environ["UVM_TOOLS_VM_ID"]
+        cls.path = "/vms/" + quote(cls.vm_id, safe="")
+        cls.vm = cls.request("GET", cls.path)
+        if (cls.vm["id"] != cls.vm_id or cls.vm["observed_status"] != "running"
+                or cls.vm["username"] == "root" or cls.vm["ports"] != []):
+            raise RuntimeError("provide a disposable running Ubuntu 22.04 VM with a non-root user and no open TCP ports")
+
+    @classmethod
+    def request(cls, method, path, body=None, *, timeout=120):
+        request = Request(cls.api_url + path, method=method,
+                          data=json.dumps(body).encode() if body is not None else None,
+                          headers={"X-UVM-Token": cls.token, "Content-Type": "application/json"})
+        with urlopen(request, timeout=timeout) as response:
+            return json.load(response)
+
+    def test_install_catalog_and_use_runtimes_as_the_guest_user(self):
+        catalog = self.request("GET", "/tools")
+        selections = [{"id": tool["id"], "version": version} for tool in catalog for version in tool["versions"] or [None]]
+        body = {"tools": selections, "request_id": str(uuid.uuid4())}
+        job = self.request("POST", self.path + "/tools/install", body)
+        deadline = time.monotonic() + 3600
+        while job["status"] in {"queued", "starting", "waiting", "installing"} and time.monotonic() < deadline:
+            time.sleep(2)
+            job = self.request("GET", self.path + "/tools/jobs/" + job["id"])
+        self.assertEqual(job["status"], "succeeded", json.dumps(job, indent=2))
+        self.assertEqual(self.request("POST", self.path + "/tools/install", body)["id"], job["id"])
+        inventory = self.request("GET", self.path + "/tools?refresh=true")
+        self.assertEqual(len(inventory["inventory"]), 12)
+        self.assertTrue(all(item["installed"] for item in inventory["inventory"]), inventory)
+
+        script = r'''set -eu
+test "$(id -u)" != 0
+git --version
+node -e 'if (1 + 1 !== 2) process.exit(1)'
+npm --version
+wget --version >/dev/null
+curl --version >/dev/null
+htop --version
+php -r 'if (2 + 2 !== 4) exit(1);'
+work=$(mktemp -d "$HOME/nest-tools-smoke.XXXXXX")
+name=$(basename "$work")
+export MAMBA_ROOT_PREFIX="$HOME/micromamba"
+cleanup() {
+    conda env remove -n "$name" -y >/dev/null 2>&1 || true
+    micromamba env remove -n "$name" -y >/dev/null 2>&1 || true
+    rm -rf -- "$work"
+}
+trap cleanup EXIT
+printf 'class Hello { public static void main(String[] args) { System.out.println("ok"); } }\n' > "$work/Hello.java"
+javac "$work/Hello.java"
+test "$(java -cp "$work" Hello)" = ok
+for version in 3.11 3.12 3.13; do
+    "python$version" -m venv "$work/venv-$version"
+    "$work/venv-$version/bin/python" -c 'import ssl, sqlite3, sys; assert sys.prefix != sys.base_prefix'
+    "$work/venv-$version/bin/python" -m pip --version
+done
+conda create -y -n "$name" python=3.13
+conda run -n "$name" python -c 'import sys; assert sys.version_info[:2] == (3, 13)'
+eval "$(conda shell.posix hook)"
+conda activate "$name"
+python -c 'import sys; assert sys.version_info[:2] == (3, 13)'
+conda deactivate
+micromamba create -y -n "$name" -c conda-forge python=3.12
+micromamba run -n "$name" python -c 'import sys; assert sys.version_info[:2] == (3, 12)'
+eval "$(micromamba shell hook --shell posix)"
+micromamba activate "$name"
+python -c 'import sys; assert sys.version_info[:2] == (3, 12)'
+micromamba deactivate
+'''
+        result = self.request("POST", self.path + "/exec",
+                              {"command": script, "sudo": False, "timeout_seconds": 1200}, timeout=1220)
+        self.assertFalse(result["timed_out"], result)
+        self.assertEqual(result["exit_code"], 0, result)
+        after = self.request("GET", self.path)
+        for key in ("id", "guest_ip", "mac", "username", "ports", "cpu", "ram_mib", "storage_mib"):
+            self.assertEqual(after[key], self.vm[key])

tests/test_server.py

@@ -878,6 +878,10 @@ class ApiRouterTests(unittest.TestCase):
                 ("/vms/{vm_id}/ports/sync", "post"),
                 ("/vms/{vm_id}/exec", "post"),
                 ("/vms/{vm_id}/agent/install", "post"),
+                ("/tools", "get"),
+                ("/vms/{vm_id}/tools", "get"),
+                ("/vms/{vm_id}/tools/install", "post"),
+                ("/vms/{vm_id}/tools/jobs/{job_id}", "get"),
                 ("/diagnostics", "get"),
             },
         )

tests/test_sessions.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 import asyncio
 import base64
 import os
+import sqlite3
 import tempfile
 import unittest
 from pathlib import Path
@@ -13,7 +14,7 @@ from nest.server import create_app
 from nest.session_store import SessionStore
 from nest.sessions import SessionManager
 from tests.test_server import FASTAPI_AVAILABLE, fake_application
-from uvm.errors import ConflictError, GuestAgentUnavailableError, GuestExecutionError, NotFoundError
+from uvm.errors import ConflictError, GuestAgentUnavailableError, GuestExecutionError, NotFoundError, StateError, ValidationError
 
 
 if FASTAPI_AVAILABLE:
@@ -28,6 +29,7 @@ class FakeTerminal:
         self.input = []
         self.sizes = []
         self.closed = False
+        self.end_calls = 0
 
     async def read(self):
         event = await self.events.get()
@@ -43,6 +45,7 @@ class FakeTerminal:
         self.sizes.append((rows, cols))
 
     async def end(self):
+        self.end_calls += 1
         await self.events.put({"type": "exit", "exit_code": 0})
 
     def close(self):
@@ -133,6 +136,70 @@ class SessionStoreTests(unittest.TestCase):
             SessionStore(self.path)
         self.assertEqual(target.read_bytes(), b"keep this")
 
+    def test_rename_preserves_session_identity_state_and_output_across_reopening(self):
+        record = self.store.create("vm-a", 24, 80)
+        self.store.activate(record["id"], "worker")
+        self.store.append(record["id"], b"history")
+        before = self.store.get("vm-a", record["id"])
+        renamed = self.store.rename("vm-a", record["id"], "  Déploiement 'alpha'  ")
+        self.assertEqual(renamed["name"], "Déploiement 'alpha'")
+        for key in before.keys() - {"name", "updated_at"}:
+            self.assertEqual(renamed[key], before[key])
+        self.store.close()
+        reopened = SessionStore(self.path)
+        self.addCleanup(reopened.close)
+        saved, _start, output = reopened.read("vm-a", record["id"], 0)
+        self.assertEqual(saved["name"], renamed["name"])
+        self.assertEqual(output, b"history")
+
+    def test_invalid_session_names_do_not_change_the_record(self):
+        record = self.store.create("vm-a", 24, 80)
+        for name in ("", "   ", "a" * 81, "two\nlines", "bad\x00name", "\ud800", 123, None):
+            with self.subTest(name=repr(name)), self.assertRaises(ValidationError):
+                self.store.rename("vm-a", record["id"], name)
+            self.assertEqual(self.store.get("vm-a", record["id"]), record)
+
+    def test_rename_and_delete_cannot_cross_vm_boundaries(self):
+        record = self.store.create("vm-a", 24, 80)
+        self.store.finish(record["id"], "closed", 0)
+        for operation in (lambda: self.store.rename("vm-b", record["id"], "other"),
+                          lambda: self.store.delete("vm-b", record["id"])):
+            with self.assertRaises(NotFoundError):
+                operation()
+        self.assertEqual(self.store.get("vm-a", record["id"])["name"], record["name"])
+
+    def test_delete_removes_metadata_and_transcript_and_keeps_other_sessions(self):
+        first = self.store.create("vm-a", 24, 80)
+        second = self.store.create("vm-a", 24, 80)
+        self.store.append(first["id"], b"remove")
+        self.store.append(second["id"], b"keep")
+        self.store.finish(first["id"], "closed", 0)
+        self.store.delete("vm-a", first["id"])
+        with self.assertRaises(NotFoundError):
+            self.store.read("vm-a", first["id"], 0)
+        with sqlite3.connect(self.path) as db:
+            self.assertEqual(db.execute("SELECT COUNT(*) FROM output WHERE session_id=?", (first["id"],)).fetchone()[0], 0)
+        self.assertEqual(self.store.read("vm-a", second["id"], 0)[2], b"keep")
+
+    def test_failed_delete_rolls_back_metadata_and_output_together(self):
+        record = self.store.create("vm-a", 24, 80)
+        self.store.append(record["id"], b"keep")
+        self.store.finish(record["id"], "closed", 0)
+        with sqlite3.connect(self.path) as db:
+            db.execute("CREATE TRIGGER reject_delete BEFORE DELETE ON sessions BEGIN SELECT RAISE(ABORT, 'test failure'); END")
+        with self.assertRaises(StateError):
+            self.store.delete("vm-a", record["id"])
+        self.assertEqual(self.store.read("vm-a", record["id"], 0)[2], b"keep")
+
+    def test_running_history_requires_closure_and_default_names_avoid_existing_labels(self):
+        records = [self.store.create("vm-a", 24, 80) for _ in range(3)]
+        with self.assertRaises(ConflictError):
+            self.store.delete("vm-a", records[1]["id"])
+        self.store.finish(records[1]["id"], "closed", 0)
+        self.store.delete("vm-a", records[1]["id"])
+        created = self.store.create("vm-a", 24, 80)
+        self.assertNotIn(created["name"], {records[0]["name"], records[2]["name"]})
+
 
 class SessionManagerTests(unittest.IsolatedAsyncioTestCase):
     async def asyncSetUp(self):
@@ -219,6 +286,83 @@ class SessionManagerTests(unittest.IsolatedAsyncioTestCase):
         self.addCleanup(reopened.close)
         self.assertEqual(reopened.read("vm-test", record["id"], 0)[2], b"guest$ ")
 
+    async def test_rename_keeps_the_same_live_terminal(self):
+        record = await self.manager.create("vm-test", 24, 80)
+        terminal = self.application.terminals[0]
+        renamed = await self.manager.rename("vm-test", record["id"], "Deployment")
+        self.assertEqual((renamed["id"], renamed["status"], renamed["name"]), (record["id"], "active", "Deployment"))
+        await self.manager.write("vm-test", record["id"], b"pwd\r")
+        self.assertEqual(terminal.input, [b"pwd\r"])
+        self.assertFalse(terminal.closed)
+        self.assertEqual(len(self.application.terminals), 1)
+
+    async def test_delete_closes_only_its_live_terminal_and_removes_all_history(self):
+        record = await self.manager.create("vm-test", 24, 80)
+        other = await self.manager.create("vm-test", 24, 80)
+        await self.manager.write("vm-test", record["id"], b"last output\r")
+        await self.manager.delete("vm-test", record["id"])
+        self.assertTrue(self.application.terminals[0].closed)
+        self.assertFalse(self.application.terminals[1].closed)
+        self.assertEqual([item["id"] for item in await self.manager.list("vm-test")], [other["id"]])
+        with sqlite3.connect(self.manager._history().path) as db:
+            self.assertEqual(db.execute("SELECT COUNT(*) FROM output WHERE session_id=?", (record["id"],)).fetchone()[0], 0)
+
+    async def test_delete_waits_for_an_inflight_launch_before_closing_its_channel(self):
+        import threading
+        entered, release = threading.Event(), threading.Event()
+        connect = self.application.lifecycle.open_terminal
+
+        def slow_connect(*args, **kwargs):
+            entered.set()
+            release.wait(2)
+            return connect(*args, **kwargs)
+
+        self.application.lifecycle.open_terminal = slow_connect
+        creating = asyncio.create_task(self.manager.create("vm-test", 24, 80))
+        self.assertTrue(await asyncio.to_thread(entered.wait, 2))
+        record = (await self.manager.list("vm-test"))[0]
+        deleting = asyncio.create_task(self.manager.delete("vm-test", record["id"]))
+        try:
+            await asyncio.sleep(0.03)
+            self.assertFalse(deleting.done())
+        finally:
+            release.set()
+            await asyncio.gather(creating, deleting)
+        self.assertTrue(self.application.terminals[0].closed)
+        self.assertEqual(await self.manager.list("vm-test"), [])
+
+    async def test_cancelled_delete_request_finishes_cleanup_even_during_shutdown(self):
+        record = await self.manager.create("vm-test", 24, 80)
+        terminal = self.application.terminals[0]
+        entered, release = asyncio.Event(), asyncio.Event()
+        end = terminal.end
+
+        async def slow_end():
+            entered.set()
+            await release.wait()
+            await end()
+
+        terminal.end = slow_end
+        request = asyncio.create_task(self.manager.delete("vm-test", record["id"]))
+        await asyncio.wait_for(entered.wait(), timeout=2)
+        request.cancel()
+        await asyncio.gather(request, return_exceptions=True)
+        release.set()
+        await self.manager.shutdown()
+        reopened = SessionStore(Path(self.temporary.name) / "nest" / "sessions.sqlite3")
+        self.addCleanup(reopened.close)
+        self.assertTrue(terminal.closed)
+        self.assertEqual(reopened.list("vm-test"), [])
+
+    async def test_close_and_delete_are_serialized_for_the_same_session(self):
+        record = await self.manager.create("vm-test", 24, 80)
+        results = await asyncio.gather(self.manager.end("vm-test", record["id"]),
+                                       self.manager.delete("vm-test", record["id"]), return_exceptions=True)
+        self.assertTrue(all(not isinstance(result, Exception) or isinstance(result, NotFoundError) for result in results))
+        self.assertTrue(self.application.terminals[0].closed)
+        self.assertEqual(self.application.terminals[0].end_calls, 1)
+        self.assertEqual(await self.manager.list("vm-test"), [])
+
 
 @unittest.skipUnless(FASTAPI_AVAILABLE, "FastAPI and httpx are required")
 class SessionApiTests(unittest.TestCase):
@@ -252,7 +396,8 @@ class SessionApiTests(unittest.TestCase):
 
     def test_all_http_operations_and_websocket_output_require_authentication(self):
         for method, path in (("GET", self.base), ("POST", self.base),
-                             ("GET", self.base + "/session-unknown"), ("POST", self.base + "/session-unknown/close")):
+                             ("GET", self.base + "/session-unknown"), ("POST", self.base + "/session-unknown/close"),
+                             ("PATCH", self.base + "/session-unknown"), ("DELETE", self.base + "/session-unknown")):
             self.assertEqual(self.client.request(method, path, json={}).status_code, 401)
         for token in ("wrong", "", "\u00e9"):
             with self.client.websocket_connect(self.base + "/session-unknown/terminal") as ws:
@@ -281,6 +426,8 @@ class SessionApiTests(unittest.TestCase):
         self.assertEqual(self.client.get(other, headers=self.headers).json(), [])
         self.assertEqual(self.client.get(other + "/" + record["id"], headers=self.headers).status_code, 404)
         self.assertEqual(self.client.post(other + "/" + record["id"] + "/close", headers=self.headers).status_code, 404)
+        self.assertEqual(self.client.patch(other + "/" + record["id"], headers=self.headers, json={"name": "Other"}).status_code, 404)
+        self.assertEqual(self.client.delete(other + "/" + record["id"], headers=self.headers).status_code, 404)
         ws = self.connect(record, vm_id="vm-other")
         self.assertEqual(ws.receive_json()["type"], "error")
         self.assertFalse(self.application.terminals[0].closed)
@@ -320,3 +467,36 @@ class SessionApiTests(unittest.TestCase):
         self.assertEqual(self.application.terminals[0].input, [])
         self.assertFalse(self.application.terminals[0].closed)
         self.assertEqual(self.client.get(self.base + "/" + record["id"], headers=self.headers).json()["status"], "active")
+
+    def test_rename_is_validated_persisted_and_sent_to_attached_terminals(self):
+        record = self.create()
+        url = self.base + "/" + record["id"]
+        for body in ({}, {"name": " "}, {"name": 123}, {"name": "x" * 81},
+                     {"name": "two\nlines"}, {"name": "bad\x00name"}, {"name": "New", "vm_id": "vm-other"}):
+            with self.subTest(body=body):
+                self.assertEqual(self.client.patch(url, headers=self.headers, json=body).status_code, 422)
+        ws = self.connect(record)
+        self.output(ws)
+        response = self.client.patch(url, headers=self.headers, json={"name": "  Déploiement  "})
+        self.assertEqual(response.status_code, 200, response.text)
+        self.assertEqual(response.json()["name"], "Déploiement")
+        while (message := ws.receive_json()).get("session", {}).get("name") != "Déploiement":
+            self.assertEqual(message["type"], "session")
+        self.assertEqual(message["session"]["status"], "active")
+        self.assertEqual(self.client.get(url, headers=self.headers).json()["name"], "Déploiement")
+        self.assertFalse(self.application.terminals[0].closed)
+
+    def test_delete_live_session_returns_no_content_and_notifies_attached_browser(self):
+        record = self.create()
+        ws = self.connect(record)
+        self.output(ws)
+        url = self.base + "/" + record["id"]
+        response = self.client.delete(url, headers=self.headers)
+        self.assertEqual(response.status_code, 204, response.text)
+        self.assertEqual(response.content, b"")
+        while (message := ws.receive_json())["type"] == "session":
+            pass
+        self.assertEqual(message, {"type": "deleted", "session_id": record["id"]})
+        self.assertTrue(self.application.terminals[0].closed)
+        self.assertEqual(self.client.get(url, headers=self.headers).status_code, 404)
+        self.assertEqual(self.client.get(self.base, headers=self.headers).json(), [])

tests/test_tool_api.py

@@ -0,0 +1,113 @@
+from __future__ import annotations
+
+import tempfile
+import unittest
+from dataclasses import replace
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import Mock
+
+from tests.test_server import FASTAPI_AVAILABLE, FakeLifecycle
+from uvm.config import Settings
+from uvm.server import create_api
+from uvm.state import StateStore
+from uvm.tools import ToolService
+
+if FASTAPI_AVAILABLE:
+    from fastapi.testclient import TestClient
+
+
+@unittest.skipUnless(FASTAPI_AVAILABLE, "FastAPI and httpx are required")
+class ToolApiTests(unittest.TestCase):
+    def setUp(self):
+        temporary = self.enterContext(tempfile.TemporaryDirectory())
+        settings = replace(Settings(), base=Path(temporary), api_token="test-token")
+        state = StateStore(settings)
+        self.lifecycle = FakeLifecycle()
+        self.lifecycle.vm.runtime_owned = True
+        settings.vm_dir(self.lifecycle.vm.id).mkdir(parents=True, mode=0o700)
+        with state.transaction() as document:
+            document.vms[self.lifecycle.vm.id] = self.lifecycle.vm
+        self.tools = ToolService(settings, state, self.lifecycle)
+        self.tools.start = Mock()
+        self.tools.shutdown = Mock()
+        self.tools.notify = Mock()
+        self.application = SimpleNamespace(settings=settings, state_store=state, lifecycle=self.lifecycle, tools=self.tools)
+        self.api = create_api(self.application, host="127.0.0.1")
+        self.client = TestClient(self.api)
+        self.addCleanup(self.client.close)
+        self.headers = {"X-UVM-Token": "test-token"}
+
+    def job(self):
+        with self.application.state_store.operation_lock():
+            return self.tools.store.enqueue_locked("vm-test", [{"id": "git"}], request_id="request-1")
+
+    def test_catalog_and_job_routes_require_the_management_token(self):
+        for method, path in (("GET", "/tools"), ("GET", "/vms/vm-test/tools"),
+                             ("POST", "/vms/vm-test/tools/install"),
+                             ("GET", "/vms/vm-test/tools/jobs/tools-missing")):
+            with self.subTest(path=path):
+                response = self.client.request(method, path, json={"tools": [{"id": "git"}]})
+                self.assertEqual(response.status_code, 401)
+        self.tools.start.assert_not_called()
+
+    def test_catalog_has_python_options_and_create_passes_normalized_selections(self):
+        result = self.client.get("/tools", headers=self.headers)
+        self.assertEqual(result.status_code, 200)
+        python = next(item for item in result.json() if item["id"] == "python")
+        self.assertEqual(python["versions"], ["3.11", "3.12", "3.13"])
+        created = self.client.post("/vms", headers=self.headers, json={"tools": [
+            {"id": "python", "version": "3.12"}, {"id": "git"}, {"id": "git"},
+        ]})
+        self.assertEqual(created.status_code, 201, created.text)
+        self.assertEqual([item.key for item in self.lifecycle.last_spec.tools], ["git", "python-3.12"])
+        self.tools.start.assert_called_once()
+        self.tools.notify.assert_called_once()
+        self.assertNotIn("password", created.json())
+
+    def test_invalid_selections_and_options_are_rejected_before_submission(self):
+        self.tools.submit = Mock()
+        invalid = [
+            {"tools": []}, {"tools": None}, {"tools": ["git"]}, {"tools": [{"id": "unknown"}]},
+            {"tools": [{"id": "python", "version": "3.14"}]}, {"tools": [{"id": "python"}]},
+            {"tools": [{"id": "git", "script": "id"}]}, {"tools": [{"id": "git"}], "start_if_stopped": "true"},
+            {"tools": [{"id": "git"}], "restart": 1}, {"tools": [{"id": "git"}], "request_id": "../../x"},
+            {"tools": [{"id": "git"}], "sudo": False},
+        ]
+        for body in invalid:
+            with self.subTest(body=body):
+                response = self.client.post("/vms/vm-test/tools/install", headers=self.headers, json=body)
+                self.assertEqual(response.status_code, 422, response.text)
+        self.tools.submit.assert_not_called()
+        for body in ({"tools": None}, {"tools": [{"id": "not-in-catalog"}]}):
+            response = self.client.post("/vms", headers=self.headers, json=body)
+            self.assertEqual(response.status_code, 422)
+        self.assertIsNone(self.lifecycle.last_spec)
+
+    def test_install_returns_accepted_job_and_whitelists_its_response(self):
+        job = self.job()
+        job.update(password="secret", host_path="/host/private")
+        self.tools.submit = Mock(return_value=job)
+        result = self.client.post("/vms/vm-test/tools/install", headers=self.headers,
+                                  json={"tools": [{"id": "git"}], "start_if_stopped": True, "request_id": "request-1"})
+        self.assertEqual(result.status_code, 202, result.text)
+        self.assertEqual(result.json()["id"], job["id"])
+        self.assertNotIn("password", result.json())
+        self.assertNotIn("host_path", result.json())
+        self.assertTrue(self.tools.submit.call_args.kwargs["start_if_stopped"])
+        self.assertEqual(self.tools.submit.call_args.args[0], "vm-test")
+
+    def test_job_history_remains_exact_vm_scoped(self):
+        job = self.job()
+        result = self.client.get(f"/vms/vm-test/tools/jobs/{job['id']}", headers=self.headers)
+        self.assertEqual(result.status_code, 200, result.text)
+        self.assertEqual(result.json()["items"][0]["status"], "queued")
+        for identifier in ("test-vm", "10.42.0.2", "vm-other"):
+            response = self.client.get(f"/vms/{identifier}/tools/jobs/{job['id']}", headers=self.headers)
+            self.assertEqual(response.status_code, 404)
+
+    def test_server_lifespan_starts_and_stops_workers(self):
+        with TestClient(self.api) as client:
+            self.tools.start.assert_called_once()
+            self.assertEqual(client.get("/tools", headers=self.headers).status_code, 200)
+        self.tools.shutdown.assert_called_once()

tests/test_tool_catalog.py

@@ -0,0 +1,162 @@
+from __future__ import annotations
+
+import ast
+import hashlib
+import io
+import shlex
+import tarfile
+import tempfile
+import unittest
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import patch
+
+from uvm import guest_tools
+from uvm.errors import ValidationError
+from uvm.tool_catalog import CATALOG, ToolSelection, all_selections, normalize_tools, recipe
+from uvm.tools import guest_command
+
+
+class ToolCatalogTests(unittest.TestCase):
+    def test_requested_catalog_and_every_guest_payload_fit_the_agent_protocol(self):
+        self.assertEqual({tool["id"] for tool in CATALOG},
+                         {"git", "nodejs", "python", "jdk", "wget", "curl", "htop", "php", "conda", "micromamba"})
+        for arch in ("x86_64", "aarch64"):
+            recipes = [recipe(item, arch) for item in all_selections()]
+            command = guest_command({"action": "probe", "tools": recipes, "username": "worker", "architecture": arch})
+            self.assertLess(len(command.encode()), 65536)
+            argv = shlex.split(command)
+            self.assertEqual(argv[:3], ["/usr/bin/python3", "-I", "-c"])
+            compile(argv[3], "guest-tools", "exec")
+            ast.parse(argv[3], feature_version=(3, 10))
+            for item in recipes:
+                if "url" in item:
+                    self.assertTrue(item["url"].startswith("https://"))
+                    self.assertNotIn("/latest", item["url"])
+                    self.assertRegex(item["sha256"], r"^[0-9a-f]{64}$")
+        self.assertEqual(len(all_selections()), 12)
+
+    def test_selection_validation_is_allowlisted_strict_and_deduplicated(self):
+        expected = (ToolSelection("git"), ToolSelection("python", "3.12"))
+        self.assertEqual(normalize_tools([{"id": "python", "version": "3.12"}, {"id": "git"}, {"id": "git"}]), expected)
+        for value in (None, "git", [], ["git"], [{"id": "git; id"}], [{"id": "python"}],
+                      [{"id": "python", "version": "3.14"}], [{"id": "git", "version": "latest"}],
+                      [{"id": "git", "command": "id"}], [{"id": ["git"]}], [{"id": "git"}] * 13):
+            with self.subTest(value=value), self.assertRaises(ValidationError):
+                normalize_tools(value)
+        self.assertEqual(normalize_tools([], allow_empty=True), ())
+
+    def test_versioned_python_commands_and_jdk_compiler_are_verified(self):
+        for version in ("3.11", "3.12", "3.13"):
+            item = recipe(ToolSelection("python", version), "aarch64")
+            self.assertEqual(set(item["links"]), {"python" + version})
+            self.assertIn("venv", item["checks"][0][-1])
+            self.assertEqual(item["checks"][1][:3], ["python" + version, "-m", "pip"])
+        jdk = recipe(ToolSelection("jdk"), "aarch64")
+        self.assertEqual(jdk["checks"], [["javac", "-version"], ["java", "-version"]])
+        self.assertIn("arm64", jdk["java_home"])
+
+
+class GuestToolRecipeTests(unittest.TestCase):
+    def setUp(self):
+        self.temporary = self.enterContext(tempfile.TemporaryDirectory())
+        self.root = Path(self.temporary)
+
+    def test_download_mismatch_removes_artifact_and_https_redirect_is_checked(self):
+        target = self.root / "download"
+        response = io.BytesIO(b"publisher payload")
+        response.geturl = lambda: "https://publisher.test/tool"
+        with patch.object(guest_tools.urllib.request, "urlopen", return_value=response):
+            with self.assertRaisesRegex(guest_tools.ToolError, "checksum"):
+                guest_tools.download("https://publisher.test/tool", target, "0" * 64)
+        self.assertFalse(target.exists())
+        response = io.BytesIO(b"publisher payload")
+        response.geturl = lambda: "https://publisher.test/tool"
+        with patch.object(guest_tools.urllib.request, "urlopen", return_value=response):
+            guest_tools.download("https://publisher.test/tool", target, hashlib.sha256(b"publisher payload").hexdigest())
+        self.assertEqual(target.read_bytes(), b"publisher payload")
+        with self.assertRaisesRegex(guest_tools.ToolError, "HTTPS"):
+            guest_tools.download("http://publisher.test/tool", target)
+        response = io.BytesIO(b"publisher payload")
+        response.geturl = lambda: "http://publisher.test/tool"
+        with patch.object(guest_tools.urllib.request, "urlopen", return_value=response):
+            with self.assertRaisesRegex(guest_tools.ToolError, "HTTPS"):
+                guest_tools.download("https://publisher.test/tool", target)
+
+    def test_archive_traversal_links_and_devices_cannot_escape_staging(self):
+        cases = [("../outside", tarfile.REGTYPE, ""), ("/outside", tarfile.REGTYPE, ""),
+                 ("link", tarfile.SYMTYPE, "../../outside"), ("link", tarfile.LNKTYPE, "../outside"),
+                 ("device", tarfile.CHRTYPE, "")]
+        for name, kind, link in cases:
+            with self.subTest(name=name, kind=kind):
+                archive = self.root / "archive.tar"
+                with tarfile.open(archive, "w") as stream:
+                    item = tarfile.TarInfo(name)
+                    item.type = kind
+                    item.linkname = link
+                    stream.addfile(item)
+                output = self.root / "output"
+                output.mkdir(exist_ok=True)
+                with self.assertRaises(guest_tools.ToolError):
+                    guest_tools.extract_archive(archive, output)
+                self.assertFalse((self.root / "outside").exists())
+
+    def test_internal_archive_symlinks_are_supported(self):
+        archive = self.root / "archive.tar"
+        with tarfile.open(archive, "w") as stream:
+            item = tarfile.TarInfo("python/bin/python3.12")
+            item.size = 4
+            item.mode = 0o755
+            stream.addfile(item, io.BytesIO(b"test"))
+            link = tarfile.TarInfo("python/bin/python3")
+            link.type = tarfile.SYMTYPE
+            link.linkname = "python3.12"
+            stream.addfile(link)
+        output = self.root / "output"
+        output.mkdir()
+        guest_tools.extract_archive(archive, output)
+        self.assertEqual((output / "python/bin/python3").read_bytes(), b"test")
+
+    def test_probe_uses_guest_account_and_rejects_wrong_runtime_or_missing_companion(self):
+        node = recipe(ToolSelection("nodejs"), "x86_64")
+        account = SimpleNamespace(pw_dir="/home/worker", pw_uid=1001, pw_gid=1002)
+        with (patch.object(guest_tools.pwd, "getpwnam", return_value=account),
+              patch.object(guest_tools.os, "getgrouplist", return_value=[1002, 1003]),
+              patch.object(guest_tools.shutil, "which", return_value="/usr/local/bin/node"),
+              patch.object(guest_tools.subprocess, "run") as run):
+            run.return_value = SimpleNamespace(returncode=0, stdout=b"v22.10.0\n")
+            result = guest_tools.probe(node, "worker")
+            self.assertFalse(result["installed"])
+            self.assertEqual(result["detected_version"], "22.10.0")
+            self.assertEqual(run.call_args.kwargs["user"], 1001)
+            self.assertEqual(run.call_args.kwargs["env"]["HOME"], "/home/worker")
+            run.side_effect = [SimpleNamespace(returncode=0, stdout=b"v24.21.0\n"),
+                               SimpleNamespace(returncode=1, stdout=b"broken npm")]
+            self.assertFalse(guest_tools.probe(node, "worker")["installed"])
+
+    def test_managed_files_preserve_edits_and_can_finish_interrupted_publication(self):
+        installer = guest_tools.Installer("root")
+        path = self.root / "profile"
+        # Isolated directory fixtures stand in for the root-owned guest hierarchy.
+        with (patch.object(guest_tools, "ROOT", self.root),
+              patch.object(guest_tools, "safe_directory"),
+              patch.object(guest_tools, "regular_file")):
+            installer.write_managed(path, b"initial")
+            installer.write_managed(path, b"updated")
+            self.assertEqual(path.read_bytes(), b"updated")
+            path.write_bytes(b"guest edit")
+            with self.assertRaisesRegex(guest_tools.ToolError, "customized"):
+                installer.write_managed(path, b"next")
+            self.assertEqual(path.read_bytes(), b"guest edit")
+            path.write_bytes(b"updated")
+            installer.remember(path, hashlib.sha256(b"next").hexdigest())
+            installer.write_managed(path, b"next")
+            self.assertEqual(path.read_bytes(), b"next")
+
+    def test_log_capture_retains_a_bounded_tail(self):
+        installer = guest_tools.Installer("root")
+        installer.append(b"a" * 20000)
+        installer.append(b"last failure")
+        self.assertEqual(len(installer.log), guest_tools.LOG_LIMIT)
+        self.assertTrue(installer.truncated)
+        self.assertTrue(installer.log.endswith(b"last failure"))

tests/test_tool_flow.py

@@ -0,0 +1,73 @@
+from __future__ import annotations
+
+import tempfile
+import time
+import unittest
+from pathlib import Path
+from unittest.mock import patch
+
+from tests.test_server import FASTAPI_AVAILABLE
+
+if FASTAPI_AVAILABLE:
+    from fastapi.testclient import TestClient
+    from tests.browser_tool_server import fixture_application
+    from uvm.server import create_api
+
+
+@unittest.skipUnless(FASTAPI_AVAILABLE, "FastAPI and httpx are required")
+class ToolFlowTests(unittest.TestCase):
+    def setUp(self):
+        temporary = self.enterContext(tempfile.TemporaryDirectory())
+        self.application = fixture_application(Path(temporary), delay=0.01)
+        self.enterContext(patch("uvm.tools.require_root"))
+        self.headers = {"X-UVM-Token": "browser-test-token"}
+
+    def wait_for(self, condition):
+        deadline = time.monotonic() + 5
+        while time.monotonic() < deadline:
+            if value := condition():
+                return value
+            time.sleep(0.01)
+        self.fail("background tool worker did not reach the expected state")
+
+    def test_create_and_later_install_use_real_workers_and_durable_job_responses(self):
+        with TestClient(create_api(self.application, host="127.0.0.1")) as client:
+            response = client.post("/vms", headers=self.headers, json={"name": "tools-vm", "tools": [
+                {"id": "git"}, {"id": "python", "version": "3.12"},
+            ]})
+            self.assertEqual(response.status_code, 201, response.text)
+            vm_id = response.json()["id"]
+            self.wait_for(lambda: self.application.tools.store.read(vm_id)["jobs"][-1]["status"] == "succeeded")
+            snapshot = client.get(f"/vms/{vm_id}/tools?refresh=true", headers=self.headers)
+            self.assertEqual(snapshot.status_code, 200, snapshot.text)
+            installed = {item["key"] for item in snapshot.json()["inventory"] if item["installed"]}
+            self.assertEqual(installed, {"git", "python-3.12", "curl"})
+            self.assertEqual(snapshot.json()["latest_job"]["source"], "create")
+            later = client.post("/vms/vm-other/tools/install", headers=self.headers,
+                                json={"tools": [{"id": "htop"}], "start_if_stopped": True})
+            self.assertEqual(later.status_code, 202, later.text)
+            job_id = later.json()["id"]
+            self.wait_for(lambda: self.application.tools.store.get_job("vm-other", job_id)["status"] == "succeeded")
+            self.assertEqual(self.application.state_store.get("vm-other").status, "running")
+            self.assertEqual(client.get(f"/vms/{vm_id}/tools/jobs/{job_id}", headers=self.headers).status_code, 404)
+
+    def test_queued_jobs_are_dispatched_on_server_start(self):
+        with self.application.state_store.operation_lock():
+            job = self.application.tools.store.enqueue_locked("vm-test", [{"id": "git"}])
+        with TestClient(create_api(self.application, host="127.0.0.1")):
+            self.wait_for(lambda: self.application.tools.store.get_job("vm-test", job["id"])["status"] == "succeeded")
+        self.assertEqual(self.application.lifecycle.attempts[("vm-test", "git")], 1)
+
+    def test_shutdown_wakes_command_waiter_and_keeps_an_interrupted_job(self):
+        self.application.lifecycle.delay = 30
+        with TestClient(create_api(self.application, host="127.0.0.1")) as client:
+            response = client.post("/vms/vm-test/tools/install", headers=self.headers, json={"tools": [{"id": "git"}]})
+            job_id = response.json()["id"]
+            self.wait_for(lambda: self.application.tools.store.get_job("vm-test", job_id)["status"] == "installing"
+                          and bool(self.application.tools._sessions))
+            started = time.monotonic()
+        self.assertLess(time.monotonic() - started, 3)
+        job = self.application.tools.store.get_job("vm-test", job_id)
+        self.assertEqual(job["status"], "interrupted")
+        self.assertTrue(job["outcome_unknown"])
+        self.assertTrue(all(not thread.is_alive() for thread in self.application.tools._threads))

tests/test_tool_jobs.py

@@ -0,0 +1,149 @@
+from __future__ import annotations
+
+import json
+import tempfile
+import unittest
+from concurrent.futures import ThreadPoolExecutor
+from dataclasses import replace
+from pathlib import Path
+from unittest.mock import Mock, patch
+
+from tests.test_server import FakeLifecycle
+from uvm.config import Settings
+from uvm.errors import ConflictError, GuestExecutionError, NotFoundError, StateError
+from uvm.state import StateStore
+from uvm.tool_catalog import ToolSelection
+from uvm.tool_store import ToolJobStore
+from uvm.tools import ToolService
+
+
+class ToolJobTests(unittest.TestCase):
+    def setUp(self):
+        temporary = self.enterContext(tempfile.TemporaryDirectory())
+        self.settings = replace(Settings(), base=Path(temporary), tool_workers=1)
+        self.state = StateStore(self.settings)
+        self.vm = replace(FakeLifecycle().vm, runtime_owned=True, pid=123, process_start_time="1", boot_id="host-boot")
+        self.settings.vm_dir(self.vm.id).mkdir(parents=True, mode=0o700)
+        with self.state.transaction() as state:
+            state.vms[self.vm.id] = self.vm
+        self.store = ToolJobStore(self.settings, self.state)
+
+    def enqueue(self, selections=None, **options):
+        with self.state.operation_lock():
+            return self.store.enqueue_locked(self.vm.id, selections or [ToolSelection("git")], **options)
+
+    def test_duplicate_concurrent_submissions_have_one_durable_job(self):
+        with ThreadPoolExecutor(max_workers=4) as executor:
+            jobs = list(executor.map(lambda _: self.enqueue(request_id="request-1"), range(4)))
+        self.assertEqual(len({job["id"] for job in jobs}), 1)
+        self.assertEqual(len(self.store.read(self.vm.id)["jobs"]), 1)
+        path = self.settings.vm_dir(self.vm.id) / "tools.json"
+        self.assertEqual(path.stat().st_mode & 0o777, 0o600)
+        self.assertNotIn("guest-secret", path.read_text())
+        with self.assertRaises(ConflictError):
+            self.enqueue([ToolSelection("htop")])
+
+    def test_history_is_vm_scoped_and_deletion_is_never_undone(self):
+        job = self.enqueue()
+        with self.assertRaises(NotFoundError):
+            self.store.get_job(self.vm.name, job["id"])
+        with self.state.transaction() as state:
+            del state.vms[self.vm.id]
+        with self.assertRaises(NotFoundError):
+            self.store.update_job(self.vm.id, job["id"], lambda current, data: current.update(status="succeeded"))
+        self.assertNotIn(self.vm.id, self.state.load().vms)
+
+    def test_recovery_preserves_queued_work_and_marks_inflight_outcomes_unknown(self):
+        job = self.enqueue()
+        self.store.recover()
+        self.assertEqual(self.store.get_job(self.vm.id, job["id"])["status"], "queued")
+        self.store.claim()
+        self.store.update_job(self.vm.id, job["id"], lambda current, data: current["items"][0].update(status="installing"))
+        self.store.recover()
+        recovered = self.store.get_job(self.vm.id, job["id"])
+        self.assertEqual(recovered["status"], "interrupted")
+        self.assertTrue(recovered["outcome_unknown"])
+        self.assertEqual(recovered["items"][0]["status"], "interrupted")
+        self.assertIsNone(self.store.claim())
+
+    def test_symlinks_and_foreign_history_are_rejected(self):
+        outside = self.settings.base / "outside"
+        outside.write_text("private data")
+        path = self.settings.vm_dir(self.vm.id) / "tools.json"
+        path.symlink_to(outside)
+        with self.assertRaises((StateError, OSError)):
+            self.store.read(self.vm.id)
+        self.assertEqual(outside.read_text(), "private data")
+        path.unlink()
+        self.enqueue()
+        data = json.loads(path.read_text())
+        data["jobs"][0]["vm_id"] = "vm-other"
+        path.write_text(json.dumps(data))
+        with self.assertRaises(StateError):
+            self.store.read(self.vm.id)
+
+    def test_only_one_worker_can_own_this_registry_and_shutdown_releases_it(self):
+        first = ToolService(self.settings, self.state, Mock())
+        second = ToolService(self.settings, self.state, Mock())
+        self.addCleanup(first.shutdown)
+        self.addCleanup(second.shutdown)
+        first.start()
+        with self.assertRaises(ConflictError):
+            second.start()
+        first.shutdown()
+        second.start()
+
+    def make_service(self):
+        lifecycle = Mock()
+        lifecycle.prepare_tool_install.return_value = self.vm
+        service = ToolService(self.settings, self.state, lifecycle)
+        return service
+
+    @staticmethod
+    def observation(selection, installed=False):
+        return {**selection.to_dict(), "key": selection.key, "installed": installed,
+                "detected_version": "2.0.0" if installed else None, "usage": "tool --version", "error": None}
+
+    def test_worker_reports_partial_failure_and_retains_successful_tools(self):
+        selections = [ToolSelection("git"), ToolSelection("htop")]
+        job = self.enqueue(selections)
+        claimed = self.store.claim()
+        service = self.make_service()
+        results = [
+            {"ok": True, "busy": False, "boot_id": "guest-boot", "tools": [self.observation(item) for item in selections]},
+            {"ok": True, "status": "installed", "tool": self.observation(selections[0], True), "output": "git installed"},
+            {"ok": False, "error": "mirror unavailable", "output": "apt download failed"},
+        ]
+        with patch.object(service, "_execute", side_effect=results) as execute:
+            service._run_job(claimed)
+        result = self.store.get_job(self.vm.id, job["id"])
+        self.assertEqual(result["status"], "failed")
+        self.assertEqual([item["status"] for item in result["items"]], ["installed", "failed"])
+        self.assertEqual(result["items"][1]["output"], "apt download failed")
+        self.assertEqual(self.state.get(self.vm.id).status, "running")
+        self.assertEqual(execute.call_args.args[1]["boot_id"], "guest-boot")
+
+    def test_lost_result_is_never_replayed_and_remaining_tools_are_skipped(self):
+        selections = [ToolSelection("git"), ToolSelection("htop")]
+        job = self.enqueue(selections)
+        service = self.make_service()
+        with patch.object(service, "_execute", side_effect=[
+            {"ok": True, "busy": False, "boot_id": "boot", "tools": [self.observation(item) for item in selections]},
+            GuestExecutionError("connection lost", outcome_unknown=True),
+        ]) as execute:
+            service._run_job(self.store.claim())
+        result = self.store.get_job(self.vm.id, job["id"])
+        self.assertEqual(execute.call_count, 2)
+        self.assertEqual(result["status"], "interrupted")
+        self.assertTrue(result["outcome_unknown"])
+        self.assertEqual([item["status"] for item in result["items"]], ["interrupted", "skipped"])
+
+    def test_prior_guest_installation_blocks_a_retry_before_submission(self):
+        job = self.enqueue()
+        service = self.make_service()
+        with patch.object(service, "_execute", return_value={"ok": True, "busy": True}) as execute:
+            service._run_job(self.store.claim())
+        self.assertEqual(execute.call_count, 1)
+        result = self.store.get_job(self.vm.id, job["id"])
+        self.assertEqual(result["status"], "interrupted")
+        self.assertFalse(result["outcome_unknown"])

tests/test_tool_lifecycle.py

@@ -0,0 +1,57 @@
+from __future__ import annotations
+
+from unittest.mock import Mock
+
+from tests.test_lifecycle import LifecycleTestCase
+from uvm.errors import ConflictError, NotFoundError, UvmError, ValidationError
+from uvm.tool_catalog import ToolSelection
+
+
+class ToolLifecycleTests(LifecycleTestCase):
+    def test_create_persists_tool_intent_before_boot(self):
+        def configure(config):
+            vm_id = next(iter(self.store.load().vms))
+            jobs = self.service._tool_store._read(vm_id)["jobs"]
+            self.assertEqual(jobs[0]["selections"], [{"id": "python", "version": "3.12"}])
+            self.assertEqual(jobs[0]["status"], "queued")
+        original = self.service._client_factory
+        def client(socket, timeout):
+            result = original(socket, timeout)
+            result.configure_and_start = configure
+            return result
+        self.service._client_factory = client
+        vm = self.create_vm(tools=(ToolSelection("python", "3.12"),))
+        self.assertEqual(vm.status, "running")
+        self.assertEqual(self.service._tool_store.read(vm.id)["jobs"][0]["source"], "create")
+
+    def test_invalid_tools_fail_before_reserving_a_vm_and_failed_boot_removes_intent(self):
+        with self.assertRaises(ValidationError):
+            self.create_vm(tools=({"id": "invalid"},))
+        self.assertEqual(self.store.load().vms, {})
+        self.host.failures["client.configure", None] = UvmError("failed boot")
+        with self.assertRaises(UvmError):
+            self.create_vm(tools=(ToolSelection("git"),))
+        self.assertEqual(list(self.settings.vms_dir.iterdir()), [])
+
+    def test_stopped_installation_starts_only_when_requested_and_aliases_are_rejected(self):
+        vm = self.create_vm(name="worker")
+        self.service.stop(vm.id)
+        for alias in (vm.name, vm.guest_ip):
+            with self.assertRaises(NotFoundError):
+                self.service.queue_tools(alias, [ToolSelection("git")], start_if_stopped=True)
+        with self.assertRaises(ConflictError):
+            self.service.queue_tools(vm.id, [ToolSelection("git")])
+        job = self.service.queue_tools(vm.id, [ToolSelection("git")], start_if_stopped=True)
+        self.assertEqual(self.store.get(vm.id).status, "stopped")
+        ready = self.service.prepare_tool_install(vm.id, start_if_stopped=job["start_if_stopped"])
+        self.assertEqual(ready.status, "running")
+        self.assertEqual(ready.guest_ip, vm.guest_ip)
+
+    def test_a_tool_step_cannot_reconnect_to_a_restarted_vm(self):
+        vm = self.create_vm()
+        identity = vm.pid, vm.process_start_time, vm.boot_id
+        self.service.restart(vm.id)
+        self.service._guest_connector = Mock()
+        with self.assertRaisesRegex(ConflictError, "restarted"):
+            self.service.execute(vm.id, "id", expected_identity=identity)
+        self.service._guest_connector.assert_not_called()

uvm/api_models.py

@@ -11,6 +11,7 @@ from .domain import VmRecord
 from .errors import ValidationError
 from .guest_agent import DEFAULT_TIMEOUT_SECONDS, MAX_COMMAND_BYTES, MAX_TIMEOUT_SECONDS
 from .guest_exec import validate_exec
+from .tool_catalog import MAX_SELECTIONS, ToolSelection
 
 
 class HealthResponse(BaseModel):
@@ -42,6 +43,19 @@ class PortRuleModel(BaseModel):
     protocol: Literal["TCP", "HTTP", "HTTPS"] = "TCP"
 
 
+class ToolSelectionModel(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    id: str = Field(strict=True, min_length=1, max_length=32)
+    version: str | None = Field(default=None, strict=True, max_length=16,
+                               description="Required for Python: 3.11, 3.12, or 3.13. Omit for other tools.")
+
+    @model_validator(mode="after")
+    def validate_selection(self):
+        ToolSelection(self.id, self.version)
+        return self
+
+
 class VmCreateRequest(BaseModel):
     model_config = ConfigDict(extra="forbid")
 
@@ -54,6 +68,8 @@ class VmCreateRequest(BaseModel):
     name: str | None = None
     ports: list[PortRuleModel] | None = None
     no_default_ports: bool = Field(default=False, strict=True)
+    tools: list[ToolSelectionModel] = Field(default_factory=list, max_length=MAX_SELECTIONS,
+                                          description="Optional guest tools to install asynchronously after boot")
 
 
 class VmResizeRequest(BaseModel):
@@ -149,6 +165,78 @@ class VmAgentInstallResponse(BaseModel):
     restart_required: bool = True
 
 
+class ToolCatalogResponse(BaseModel):
+    id: str
+    name: str
+    description: str
+    version_label: str
+    versions: list[str]
+    source: str
+
+
+class ToolInstallRequest(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    tools: list[ToolSelectionModel] = Field(min_length=1, max_length=MAX_SELECTIONS)
+    start_if_stopped: bool = Field(default=False, strict=True)
+    restart: bool = Field(default=False, strict=True, description="Restart first to refresh the guest agent")
+    request_id: str | None = Field(default=None, strict=True, max_length=80, pattern=r"^[A-Za-z0-9_-]{1,80}$",
+                                   description="Optional idempotency key, retained with the most recent 10 jobs")
+
+
+class ToolInventoryResponse(BaseModel):
+    key: str
+    id: str
+    version: str | None
+    installed: bool
+    detected_version: str | None
+    error: str | None
+    usage: str
+
+
+class ToolJobItemResponse(BaseModel):
+    key: str
+    id: str
+    version: str | None
+    status: Literal["queued", "installing", "installed", "already_installed", "failed", "interrupted", "skipped"]
+    detected_version: str | None
+    error: str | None
+    output: str
+    truncated: bool
+
+
+class ToolJobSummaryResponse(BaseModel):
+    id: str
+    status: Literal["queued", "starting", "waiting", "installing", "succeeded", "failed", "interrupted"]
+    created_at: float
+    completed_at: float | None
+
+
+class ToolJobResponse(ToolJobSummaryResponse):
+    vm_id: str
+    source: Literal["create", "user"]
+    updated_at: float
+    start_if_stopped: bool
+    restart: bool
+    request_id: str | None
+    error: str | None
+    outcome_unknown: bool
+    items: list[ToolJobItemResponse]
+
+
+class VmToolsResponse(BaseModel):
+    vm_id: str
+    inventory: list[ToolInventoryResponse]
+    checked_at: float | None
+    agent_ready: bool
+    guest_busy: bool
+    supported: bool | None
+    error: str | None
+    stale: bool
+    latest_job: ToolJobResponse | None
+    jobs: list[ToolJobSummaryResponse]
+
+
 class DiagnosticResponse(BaseModel):
     name: str
     ok: bool

uvm/app.py

@@ -14,6 +14,7 @@ from .lifecycle import LifecycleService
 from .network import NetworkManager
 from .state import StateStore
 from .system import CommandRunner
+from .tools import ToolService
 
 
 @dataclass(slots=True)
@@ -24,6 +25,7 @@ class Application:
     state_store: StateStore
     installer: Installer
     lifecycle: LifecycleService
+    tools: ToolService | None = None
 
 
 def build_application(emit: Callable[[str], None] | None = print) -> Application:
@@ -41,4 +43,5 @@ def build_application(emit: Callable[[str], None] | None = print) -> Application
         state_store=state_store,
         installer=installer,
         lifecycle=lifecycle,
+        tools=ToolService(settings, state_store, lifecycle),
     )

uvm/config.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import math
 import os
 from dataclasses import dataclass, field
 from ipaddress import IPv4Address, IPv4Network
@@ -46,6 +47,9 @@ class Settings:
     api_timeout_s: float = 2.0
     api_socket_timeout_s: float = 5.0
     terminate_timeout_s: float = 5.0
+    tool_workers: int = 2
+    tool_ready_timeout_s: float = 120.0
+    tool_install_timeout_s: float = 1200.0
 
     def __post_init__(self) -> None:
         architecture = self.architecture.lower()
@@ -54,6 +58,13 @@ class Settings:
             raise ConfigurationError(f"unsupported host architecture: {self.architecture}")
         object.__setattr__(self, "architecture", architecture)
 
+        if type(self.tool_workers) is not int or not 1 <= self.tool_workers <= 8:
+            raise ConfigurationError("tool_workers must be an integer between 1 and 8")
+        for name in ("tool_ready_timeout_s", "tool_install_timeout_s"):
+            value = getattr(self, name)
+            if type(value) not in (int, float) or not 0 < value <= 3600 or not math.isfinite(value):
+                raise ConfigurationError(f"{name} must be finite, greater than zero, and at most 3600 seconds")
+
         # Resolve defaults at construction time, while preserving individual URL overrides.
         asset_base = f"{DEFAULT_GUEST_ASSETS_URL}/{architecture}"
         if not self.kernel_url:

uvm/domain.py

@@ -11,6 +11,7 @@ from typing import Any
 
 from .config import MAX_STORAGE_MIB, MIN_STORAGE_MIB
 from .errors import StateError, ValidationError
+from .tool_catalog import ToolSelection
 
 
 STATE_SCHEMA_VERSION = 3
@@ -55,6 +56,7 @@ class VmSpec:
     name: str | None = None
     ports: tuple[PortRule, ...] = (PortRule(22),)
     storage_mib: int = MIN_STORAGE_MIB
+    tools: tuple[ToolSelection, ...] = ()
 
 
 @dataclass(frozen=True, slots=True)

uvm/guest_exec.py

@@ -111,4 +111,10 @@ class GuestSession:
             ) from error
 
     def close(self) -> None:
+        # Wake a thread waiting for a long-running tool command on server shutdown.
+        # The guest still owns its bounded command and package-manager cleanup.
+        try:
+            self._connection.shutdown(socket.SHUT_RDWR)
+        except OSError:
+            pass
         self._connection.close()

uvm/guest_tools.py

@@ -0,0 +1,400 @@
+"""Standalone guest-only tool installer, submitted through the execution agent.
+
+Importing this module on the host has no side effects. It uses only Python 3.10
+stdlib; all package commands, downloads, and filesystem changes run in the guest.
+"""
+
+from __future__ import annotations
+
+import fcntl
+import hashlib
+import json
+import os
+import pwd
+import re
+import shlex
+import shutil
+import stat
+import subprocess
+import sys
+import tarfile
+import tempfile
+import time
+import urllib.request
+from pathlib import Path
+
+
+ROOT = Path("/opt/uvm-tools")
+LOCK = Path("/run/uvm-tools.lock")
+BIN_DIR = Path("/usr/local/bin")
+PROFILE_DIR = Path("/etc/profile.d")
+PATH = "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
+LOG_LIMIT = 16384
+PHP_KEY = "B8DC7E53946656EFBCE4C1DD71DAEAAB4AD4CAB6"
+
+
+class ToolError(Exception):
+    pass
+
+
+def atomic_write(path, data, mode=0o644):
+    descriptor, name = tempfile.mkstemp(prefix=".uvm-tools-", dir=path.parent)
+    try:
+        with os.fdopen(descriptor, "wb") as stream:
+            os.fchmod(stream.fileno(), mode)
+            stream.write(data)
+            stream.flush()
+            os.fsync(stream.fileno())
+        os.replace(name, path)
+        directory = os.open(path.parent, os.O_RDONLY | os.O_DIRECTORY)
+        try:
+            os.fsync(directory)
+        finally:
+            os.close(directory)
+    finally:
+        Path(name).unlink(missing_ok=True)
+
+
+def safe_directory(path):
+    for parent in reversed((path, *path.parents)):
+        if not parent.exists():
+            parent.mkdir(mode=0o755)
+        metadata = parent.lstat()
+        if not stat.S_ISDIR(metadata.st_mode) or metadata.st_uid != os.geteuid() or metadata.st_mode & 0o022:
+            raise ToolError(f"tool installation needs a root-owned directory: {parent}")
+
+
+def regular_file(path):
+    metadata = path.lstat()
+    if not stat.S_ISREG(metadata.st_mode) or metadata.st_uid != os.geteuid() or metadata.st_nlink != 1 or metadata.st_mode & 0o022:
+        raise ToolError(f"refusing to replace an unmanaged tool file: {path}")
+
+
+def download(url, destination, digest=None, *, max_bytes=1024 * 1024 * 1024):
+    if not url.startswith("https://"):
+        raise ToolError("tool downloads require HTTPS")
+    checksum = hashlib.sha256()
+    size = 0
+    with urllib.request.urlopen(url, timeout=30) as response, destination.open("wb") as output:
+        if not response.geturl().startswith("https://"):
+            raise ToolError("tool download redirected away from HTTPS")
+        while True:
+            chunk = response.read(65536)
+            if not chunk:
+                break
+            size += len(chunk)
+            if size > max_bytes:
+                raise ToolError("tool download exceeds its size limit")
+            checksum.update(chunk)
+            output.write(chunk)
+    if digest is not None and checksum.hexdigest() != digest:
+        destination.unlink(missing_ok=True)
+        raise ToolError("tool download checksum did not match the pinned release")
+
+
+def extract_archive(archive_path, destination):
+    """Bound extraction and check each path, including previously extracted links."""
+    root = destination.resolve()
+    with tarfile.open(archive_path) as archive:
+        members = archive.getmembers()
+        if len(members) > 50000 or sum(member.size for member in members) > 4 * 1024**3:
+            raise ToolError("tool archive exceeds its extraction limit")
+        for member in members:
+            name = Path(member.name)
+            target = destination / name
+            if name.is_absolute() or ".." in name.parts or not target.resolve().is_relative_to(root):
+                raise ToolError("unsafe path in tool archive")
+            if not (member.isfile() or member.isdir() or member.issym() or member.islnk()):
+                raise ToolError("unsupported file in tool archive")
+            if member.issym() or member.islnk():
+                link = (target.parent if member.issym() else destination) / member.linkname
+                if Path(member.linkname).is_absolute() or not link.resolve().is_relative_to(root):
+                    raise ToolError("unsafe link in tool archive")
+            member.mode &= 0o755
+            member.uid = member.gid = 0
+            member.uname = member.gname = "root"
+            options = {"filter": "data"} if hasattr(tarfile, "data_filter") else {}
+            archive.extract(member, destination, **options)
+
+
+def platform_info():
+    values = {}
+    for line in Path("/etc/os-release").read_text().splitlines():
+        if "=" in line:
+            key, value = line.split("=", 1)
+            values[key] = value.strip('"')
+    arch = os.uname().machine
+    return {"os": values.get("PRETTY_NAME", "Unknown OS"), "architecture": arch,
+            "supported": values.get("ID") == "ubuntu" and values.get("VERSION_ID") == "22.04"
+                         and arch in ("x86_64", "aarch64"),
+            "boot_id": Path("/proc/sys/kernel/random/boot_id").read_text().strip()}
+
+
+def guest_busy():
+    try:
+        descriptor = os.open(LOCK, os.O_RDONLY | os.O_NOFOLLOW)
+    except FileNotFoundError:
+        return False
+    try:
+        try:
+            fcntl.flock(descriptor, fcntl.LOCK_EX | fcntl.LOCK_NB)
+            return False
+        except BlockingIOError:
+            return True
+    finally:
+        os.close(descriptor)
+
+
+def probe(tool, username):
+    account = pwd.getpwnam(username)
+    environment = {"PATH": PATH, "HOME": account.pw_dir, "USER": username, "LOGNAME": username,
+                   "LANG": "C.UTF-8", "LC_ALL": "C.UTF-8", "PYTHONDONTWRITEBYTECODE": "1"}
+    result = {"key": tool["key"], "id": tool["id"], "version": tool["version"], "installed": False,
+              "detected_version": None, "error": None, "usage": tool["usage"]}
+    for index, command in enumerate(tool["checks"]):
+        if shutil.which(command[0], path=PATH) is None:
+            return result
+        try:
+            process = subprocess.run(command, stdin=subprocess.DEVNULL, stdout=subprocess.PIPE,
+                                     stderr=subprocess.STDOUT, timeout=3, env=environment, cwd=account.pw_dir,
+                                     user=account.pw_uid, group=account.pw_gid,
+                                     extra_groups=os.getgrouplist(username, account.pw_gid))
+            output = process.stdout[:4096].decode("utf-8", errors="replace")
+        except (OSError, subprocess.TimeoutExpired) as error:
+            result["error"] = f"could not verify {command[0]}: {error}"[:512]
+            return result
+        if process.returncode != 0:
+            result["error"] = f"{command[0]} verification failed: {output}"[:512]
+            return result
+        match = re.search(tool["match"], output)
+        if index == 0:
+            if not match:
+                result["error"] = f"could not determine the installed {tool['id']} version"
+                return result
+            result["detected_version"] = match.group(1)
+            if not result["detected_version"].startswith(tool["prefix_match"]):
+                return result
+        elif tool["id"] == "jdk" and (not match or not match.group(1).startswith("21.")):
+            return result
+    result["installed"] = True
+    return result
+
+
+class Installer:
+    def __init__(self, username):
+        self.username = username
+        self.log = bytearray()
+        self.truncated = False
+        self.lock = None
+        self.managed = {}
+
+    def append(self, data):
+        self.log.extend(data)
+        if len(self.log) > LOG_LIMIT:
+            del self.log[:-LOG_LIMIT]
+            self.truncated = True
+
+    def run(self, command):
+        self.append(("+ " + shlex.join(map(str, command)) + "\n").encode())
+        environment = {**os.environ, "PATH": PATH, "LC_ALL": "C.UTF-8", "DEBIAN_FRONTEND": "noninteractive"}
+        with subprocess.Popen(command, stdin=subprocess.DEVNULL, stdout=subprocess.PIPE,
+                              stderr=subprocess.STDOUT, env=environment, pass_fds=(self.lock,)) as process:
+            while True:
+                chunk = process.stdout.read(8192)
+                if not chunk:
+                    break
+                self.append(chunk)
+            if process.wait() != 0:
+                raise ToolError(f"{command[0]} failed with exit code {process.returncode}; see the installation output")
+
+    def remember(self, path, value):
+        name = str(path)
+        self.managed[name] = list(dict.fromkeys([*self.managed.get(name, []), value]))[-2:]
+        atomic_write(ROOT / "managed.json", json.dumps(self.managed).encode(), 0o600)
+
+    def write_managed(self, path, data, mode=0o644):
+        safe_directory(path.parent)
+        digest = hashlib.sha256(data).hexdigest()
+        if path.exists() or path.is_symlink():
+            regular_file(path)
+            current = hashlib.sha256(path.read_bytes()).hexdigest()
+            if current == digest:
+                return
+            if current not in self.managed.get(str(path), []):
+                raise ToolError(f"customized file needs attention before installing: {path}")
+        # Record intent first so an interrupted publication can be retried.
+        self.remember(path, digest)
+        atomic_write(path, data, mode)
+
+    def link(self, name, target):
+        path = BIN_DIR / name
+        safe_directory(path.parent)
+        target = str(target)
+        if path.exists() or path.is_symlink():
+            if path.is_symlink() and os.readlink(path) == target:
+                return
+            if not path.is_symlink() or os.readlink(path) not in self.managed.get(str(path), []):
+                raise ToolError(f"customized executable needs attention before installing: {path}")
+        self.remember(path, target)
+        temporary = path.with_name(".uvm-" + name)
+        temporary.unlink(missing_ok=True)
+        temporary.symlink_to(target)
+        os.replace(temporary, path)
+
+    def apt(self, packages, *, refresh=False):
+        stamp = ROOT / "apt-updated"
+        if refresh or not stamp.exists() or time.time() - stamp.stat().st_mtime > 300:
+            self.run(["/usr/bin/apt-get", "-o", "DPkg::Lock::Timeout=300", "-o", "Acquire::Retries=2",
+                      "-o", "APT::Update::Error-Mode=any", "update"])
+            atomic_write(stamp, b"updated\n", 0o600)
+        pending = ROOT / "apt-pending"
+        if pending.exists():
+            # Complete configuration left by an interrupted tool installation.
+            self.run(["/usr/bin/dpkg", "--configure", "-a"])
+        atomic_write(pending, b"pending\n", 0o600)
+        self.run(["/usr/bin/apt-get", "-o", "DPkg::Lock::Timeout=300", "-o", "Acquire::Retries=2",
+                  "-o", "Dpkg::Options::=--force-confold", "-y", "--no-install-recommends", "install", *packages])
+        pending.unlink()
+
+    def php_repository(self, temporary):
+        self.apt(["ca-certificates", "gnupg"])
+        key = temporary / "php.asc"
+        download(f"https://keyserver.ubuntu.com/pks/lookup?op=get&search=0x{PHP_KEY}", key, max_bytes=65536)
+        result = subprocess.run(["gpg", "--batch", "--homedir", str(temporary), "--with-colons", "--show-keys", str(key)],
+                                capture_output=True, check=True)
+        fingerprints = [line.split(":")[9] for line in result.stdout.decode().splitlines() if line.startswith("fpr:")]
+        if not fingerprints or fingerprints[0] != PHP_KEY:
+            raise ToolError("PHP repository signing key did not match its pinned fingerprint")
+        binary_key = temporary / "php.gpg"
+        self.run(["gpg", "--batch", "--homedir", str(temporary), "--dearmor", "--output", str(binary_key), str(key)])
+        self.write_managed(Path("/etc/apt/keyrings/uvm-php.gpg"), binary_key.read_bytes())
+        self.write_managed(Path("/etc/apt/sources.list.d/uvm-php.list"),
+                           b"deb [signed-by=/etc/apt/keyrings/uvm-php.gpg] https://ppa.launchpadcontent.net/ondrej/php/ubuntu jammy main\n")
+
+    def payload(self, tool, temporary):
+        prefix = ROOT / tool["release"]
+        marker = prefix / ".uvm-tools.json"
+        identity = {"release": tool["release"], "sha256": tool["sha256"]}
+        if prefix.exists() or prefix.is_symlink():
+            safe_directory(prefix)
+            if not marker.exists():
+                raise ToolError(f"installation directory is already in use: {prefix}")
+            regular_file(marker)
+            previous = json.loads(marker.read_bytes())
+            if not isinstance(previous, dict) or any(previous.get(key) != value for key, value in identity.items()):
+                raise ToolError(f"installation directory belongs to a different release: {prefix}")
+            if tool["kind"] != "conda" or previous.get("complete"):
+                return prefix
+        artifact = temporary / "download"
+        self.append(f"Downloading {tool['release']} and verifying SHA-256…\n".encode())
+        download(tool["url"], artifact, tool["sha256"])
+        if tool["kind"] == "archive":
+            extraction = temporary / "extract"
+            extraction.mkdir()
+            extract_archive(artifact, extraction)
+            source = extraction / tool["archive_root"]
+            if not source.is_dir() or source.is_symlink():
+                raise ToolError("tool archive has an unexpected layout")
+            atomic_write(source / ".uvm-tools.json", json.dumps(identity).encode(), 0o600)
+            os.replace(source, prefix)
+        else:
+            prefix.mkdir(exist_ok=True, mode=0o755)
+            atomic_write(marker, json.dumps(identity).encode(), 0o600)
+            if tool["kind"] == "conda":
+                self.run(["/bin/bash", str(artifact), "-b", "-f", "-p", str(prefix)])
+            else:
+                artifact.chmod(0o755)
+                os.replace(artifact, prefix / "micromamba")
+        return prefix
+
+    def configure(self, tool, prefix):
+        for name, target in tool["links"].items():
+            self.link(name, Path(target) if target.startswith("/") else prefix / target)
+        if tool["id"] == "python":
+            executable = prefix / "bin" / f"python{tool['version']}"
+            self.write_managed(BIN_DIR / f"pip{tool['version']}",
+                               f'#!/bin/sh\nexec {shlex.quote(str(executable))} -m pip "$@"\n'.encode(), 0o755)
+        elif tool["id"] == "jdk":
+            self.write_managed(PROFILE_DIR / "uvm-java.sh",
+                               f"export JAVA_HOME={shlex.quote(tool['java_home'])}\n".encode())
+        elif tool["id"] == "conda":
+            for key, value in (("auto_activate_base", "false"),):
+                self.run([str(prefix / "bin/conda"), "config", "--system", "--set", key, value])
+            for key, value in (("envs_dirs", "~/.conda/envs"), ("pkgs_dirs", "~/.conda/pkgs")):
+                self.run([str(prefix / "bin/conda"), "config", "--system", "--prepend", key, value])
+            self.write_managed(PROFILE_DIR / "uvm-conda.sh",
+                               f". {shlex.quote(str(prefix / 'etc/profile.d/conda.sh'))}\n".encode())
+        elif tool["id"] == "micromamba":
+            self.write_managed(PROFILE_DIR / "uvm-micromamba.sh",
+                               b'export MAMBA_ROOT_PREFIX="${MAMBA_ROOT_PREFIX:-$HOME/micromamba}"\n'
+                               b'eval "$(/usr/local/bin/micromamba shell hook --shell posix)"\n')
+
+    def install(self, tool):
+        safe_directory(LOCK.parent)
+        self.lock = os.open(LOCK, os.O_RDWR | os.O_CREAT | os.O_NOFOLLOW, 0o600)
+        try:
+            metadata = os.fstat(self.lock)
+            if not stat.S_ISREG(metadata.st_mode) or metadata.st_nlink != 1 or metadata.st_uid != os.geteuid():
+                raise ToolError("unsafe guest tool installation lock")
+            try:
+                fcntl.flock(self.lock, fcntl.LOCK_EX | fcntl.LOCK_NB)
+            except BlockingIOError:
+                raise ToolError("another installation is still running inside this guest; wait before retrying")
+            existing = probe(tool, self.username)
+            if existing["installed"]:
+                return {"status": "already_installed", "tool": existing}
+            safe_directory(ROOT)
+            manifest = ROOT / "managed.json"
+            if manifest.exists():
+                regular_file(manifest)
+                self.managed = json.loads(manifest.read_bytes())
+                if (not isinstance(self.managed, dict) or any(
+                        not isinstance(values, list) or len(values) > 2 or any(not isinstance(value, str) for value in values)
+                        for values in self.managed.values())):
+                    raise ToolError("invalid managed tool file manifest")
+            with tempfile.TemporaryDirectory(prefix=".install-", dir=ROOT) as name:
+                temporary = Path(name)
+                if tool["kind"] == "php":
+                    self.php_repository(temporary)
+                self.apt(tool["packages"], refresh=tool["kind"] == "php")
+                prefix = self.payload(tool, temporary) if "url" in tool else None
+                self.configure(tool, prefix)
+                installed = probe(tool, self.username)
+                if not installed["installed"]:
+                    raise ToolError(installed["error"] or f"{tool['key']} did not pass its version check after installation")
+                if prefix is not None:
+                    atomic_write(prefix / ".uvm-tools.json", json.dumps({"release": tool["release"],
+                                 "sha256": tool["sha256"], "complete": True}).encode(), 0o600)
+                return {"status": "installed", "tool": installed}
+        finally:
+            os.close(self.lock)
+            self.lock = None
+
+
+def main(payload):
+    installer = Installer(payload["username"])
+    info = {}
+    try:
+        if os.geteuid() != 0:
+            raise ToolError("tool installation and inspection require guest root")
+        info = platform_info()
+        info["supported"] = info["supported"] and info["architecture"] == payload["architecture"]
+        if not info["supported"]:
+            raise ToolError("one-click tools require an Ubuntu 22.04 guest on the selected x86_64/ARM64 architecture")
+        if payload.get("boot_id") and payload["boot_id"] != info["boot_id"]:
+            raise ToolError("the guest rebooted during installation; retry on the current boot")
+        if payload["action"] == "probe":
+            result = {**info, "busy": guest_busy(), "tools": [probe(tool, payload["username"]) for tool in payload["tools"]]}
+        elif payload["action"] == "install":
+            result = installer.install(payload["tools"][0])
+        else:
+            raise ToolError("unsupported tool operation")
+        result["ok"] = True
+    except (ToolError, OSError, ValueError, KeyError, tarfile.TarError, subprocess.SubprocessError) as error:
+        result = {"ok": False, "error": str(error)[:2048], "supported": info.get("supported")}
+    result.update(output=installer.log.decode("utf-8", errors="replace"), truncated=installer.truncated)
+    print(json.dumps(result, ensure_ascii=True))
+
+
+if __name__ == "__main__":
+    main(json.loads(sys.argv[1]))

uvm/lifecycle.py

@@ -26,6 +26,8 @@ from .known_hosts import KnownHostsManager
 from .network import NetworkManager
 from .state import StateStore
 from .system import check_kvm, ensure_data_directories, require_root
+from .tool_catalog import normalize_tools
+from .tool_store import ToolJobStore
 from .validation import parse_name, parse_password, parse_ports, parse_timeout, parse_username
 
 
@@ -75,6 +77,7 @@ class LifecycleService:
         self._guest_connector = guest_connector
         self._terminal_connector = terminal_connector
         self._clock = clock
+        self._tool_store = ToolJobStore(settings, state_store, clock=clock)
 
     def create(self, spec: VmSpec) -> VmRecord:
         require_root()
@@ -84,6 +87,7 @@ class LifecycleService:
         if spec.name is not None:
             parse_name(spec.name)
         ports = self._validated_ports(spec.ports)
+        tools = normalize_tools(spec.tools, allow_empty=True)
         ensure_data_directories(self._settings)
         with self._state_store.operation_lock():
             self._check_boot()
@@ -100,6 +104,8 @@ class LifecycleService:
                 self._images.provision_guest_agent(disk, vm.id, vm.username)
                 vm.disk_ready = True
                 self._save(vm)
+                if tools:
+                    self._tool_store.enqueue_locked(vm.id, tools, source="create")
                 self._boot(vm, assets.kernel)
                 return vm
             except BaseException as error:
@@ -506,6 +512,8 @@ class LifecycleService:
     def execute(
         self, vm_id: str, command: str, *, sudo: bool = False,
         timeout_seconds: float = DEFAULT_TIMEOUT_SECONDS,
+        expected_identity: tuple | None = None,
+        on_connect: Callable[[GuestSession], None] | None = None,
     ) -> ExecResult:
         """Pin a running VM's guest connection under the lock, then wait unlocked."""
         require_root()
@@ -516,14 +524,52 @@ class LifecycleService:
             if (vm.status != "running" or vm.operation is not None or vm.pending_resize is not None
                     or self._process.observe(vm) != "alive"):
                 raise ConflictError("guest execution requires a running VM with no pending lifecycle operation")
+            if expected_identity is not None and expected_identity != (vm.pid, vm.process_start_time, vm.boot_id):
+                raise ConflictError("the VM restarted during tool installation; retry on the current boot")
             session = self._guest_connector(self._settings.guest_socket_path(vm.id), vm.id, vm.username, vm.pid)
         try:
+            if on_connect is not None:
+                on_connect(session)
             # This session stays attached to that VMM even if the VM is stopped,
             # deleted, or restarted concurrently. Never reconnect/retry a command.
             return session.execute(command, sudo=sudo, timeout_seconds=timeout_seconds)
         finally:
             session.close()
 
+    def _validate_tool_target(self, vm: VmRecord, *, start_if_stopped: bool, restart: bool) -> str:
+        self._not_deleting(vm)
+        observation = self._process.observe(vm)
+        if vm.operation is not None or vm.pending_resize is not None or not vm.disk_ready:
+            raise ConflictError("finish pending lifecycle operations before installing tools")
+        if vm.status == "running" and observation == "alive":
+            return "running"
+        if vm.status == "stopped" and observation == "gone" and (start_if_stopped or restart):
+            self._validate_resume(vm)
+            return "stopped"
+        raise ConflictError("tool installation requires a running VM; use Start & install for a stopped VM")
+
+    def queue_tools(self, vm_id: str, selections, *, start_if_stopped=False, restart=False, request_id=None) -> dict:
+        require_root()
+        selections = normalize_tools(selections)
+        with self._state_store.operation_lock():
+            vm = self._load_for_operation(vm_id, exact_id=True)
+            self._validate_tool_target(vm, start_if_stopped=start_if_stopped, restart=restart)
+            return self._tool_store.enqueue_locked(vm.id, selections, start_if_stopped=start_if_stopped,
+                                                   restart=restart, request_id=request_id)
+
+    def prepare_tool_install(self, vm_id: str, *, start_if_stopped=False, restart=False) -> VmRecord:
+        """Perform requested startup/agent refresh atomically before a tool worker connects."""
+        require_root()
+        with self._state_store.operation_lock():
+            vm = self._load_for_operation(vm_id, exact_id=True)
+            status = self._validate_tool_target(vm, start_if_stopped=start_if_stopped, restart=restart)
+            if status == "running" and restart:
+                self._stop_locked(vm)
+            if status == "stopped" or restart:
+                # The normal retained-disk boot also provisions the current agent.
+                vm = self._start_locked(vm)
+            return vm
+
     def open_terminal(self, vm_id: str, *, rows: int = 24, cols: int = 80) -> tuple[GuestTerminal, str]:
         """Connect to the exact running VMM under the lock, then open its guest PTY."""
         require_root()

uvm/routers/tools.py

@@ -0,0 +1,46 @@
+"""Authenticated tool catalog, installation jobs, and guest inventory."""
+
+from __future__ import annotations
+
+from fastapi import APIRouter, Depends, status
+
+from ..api_models import ToolCatalogResponse, ToolInstallRequest, ToolJobResponse, VmToolsResponse
+from ..app import Application
+from ..errors import UvmError
+from ..tool_catalog import CATALOG, normalize_tools
+from .dependencies import get_authorized_application
+
+
+router = APIRouter(tags=["tools"])
+
+
+def tool_service(application):
+    service = getattr(application, "tools", None)
+    if service is None:
+        raise UvmError("guest tool installation is unavailable in this application")
+    return service
+
+
+@router.get("/tools", response_model=list[ToolCatalogResponse])
+def catalog(application: Application = Depends(get_authorized_application)):
+    return list(CATALOG)
+
+
+@router.get("/vms/{vm_id}/tools", response_model=VmToolsResponse)
+def vm_tools(vm_id: str, refresh: bool = False, application: Application = Depends(get_authorized_application)):
+    """Return cached observations; refresh=true probes the running guest without installing."""
+    return tool_service(application).snapshot(vm_id, refresh=refresh)
+
+
+@router.post("/vms/{vm_id}/tools/install", response_model=ToolJobResponse, status_code=status.HTTP_202_ACCEPTED)
+def install_tools(vm_id: str, request: ToolInstallRequest,
+                  application: Application = Depends(get_authorized_application)):
+    return tool_service(application).submit(
+        vm_id, normalize_tools([item.model_dump() for item in request.tools]),
+        start_if_stopped=request.start_if_stopped, restart=request.restart, request_id=request.request_id,
+    )
+
+
+@router.get("/vms/{vm_id}/tools/jobs/{job_id}", response_model=ToolJobResponse)
+def tool_job(vm_id: str, job_id: str, application: Application = Depends(get_authorized_application)):
+    return tool_service(application).store.get_job(vm_id, job_id)

uvm/routers/vms.py

@@ -22,6 +22,7 @@ from ..api_models import (
 from ..app import Application
 from ..domain import PortRule, VmRecord, VmSpec
 from ..errors import ValidationError
+from ..tool_catalog import normalize_tools
 from ..validation import (
     parse_cpu,
     parse_name,
@@ -33,6 +34,7 @@ from ..validation import (
     parse_username,
 )
 from .dependencies import get_authorized_application
+from .tools import tool_service
 
 
 router = APIRouter(prefix="/vms", tags=["vms"])
@@ -73,6 +75,9 @@ def create_vm(
     ports = parse_ports([f"{rule.port}:{rule.protocol}" for rule in request.ports or ()])
     if not request.no_default_ports and not any(rule.port == 22 for rule in ports):
         ports = tuple(sorted((*ports, PortRule(22)), key=lambda rule: rule.port))
+    tools = normalize_tools([item.model_dump() for item in request.tools], allow_empty=True)
+    if tools:
+        tool_service(application).start()
     vm = application.lifecycle.create(
         VmSpec(
             cpu=parse_cpu(str(request.cpu)),
@@ -83,8 +88,11 @@ def create_vm(
             password=parse_password(request.password.get_secret_value()),
             name=parse_name(request.name) if request.name is not None else None,
             ports=ports,
+            tools=tools,
         )
     )
+    if tools:
+        tool_service(application).notify()
     return vm_response(vm, policy_status=_mutation_policy_status(vm))
 
 

uvm/server.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 from collections.abc import Callable
+from contextlib import asynccontextmanager
 from ipaddress import ip_address
 from pathlib import Path
 import ssl
@@ -32,6 +33,7 @@ def create_api(application: Application | None = None, *, host: str | None = Non
     try:
         from fastapi import FastAPI
         from fastapi.responses import JSONResponse
+        from starlette.concurrency import run_in_threadpool
     except ModuleNotFoundError as error:
         raise UvmError(
             "FastAPI server support is not installed. Install this project with its dependencies."
@@ -42,11 +44,24 @@ def create_api(application: Application | None = None, *, host: str | None = Non
     from .routers.installation import router as installation_router
     from .routers.vms import router as vms_router
     from .routers.diagnostics import router as diagnostics_router
+    from .routers.tools import router as tools_router
+
+    @asynccontextmanager
+    async def lifespan(_api):
+        tools = getattr(resolved_application, "tools", None)
+        if tools is not None:
+            await run_in_threadpool(tools.start)
+        try:
+            yield
+        finally:
+            if tools is not None:
+                await run_in_threadpool(tools.shutdown)
 
     api = FastAPI(
         title="uvm",
         version="0.1.0",
         description="Local Firecracker microVM management API.",
+        lifespan=lifespan,
     )
     api.state.uvm_application = resolved_application
 
@@ -65,6 +80,7 @@ def create_api(application: Application | None = None, *, host: str | None = Non
     api.include_router(installation_router)
     api.include_router(vms_router)
     api.include_router(diagnostics_router)
+    api.include_router(tools_router)
 
     return api
 

uvm/tool_catalog.py

@@ -0,0 +1,162 @@
+"""The allowlisted guest tool catalog and pinned, architecture-specific recipes.
+
+Download digests come from the publishers' release SHA256SUMS/.sha256 files:
+Node v24.21.0, python-build-standalone 20260924, Miniforge 26.5.3-0, and
+micromamba-releases 2.9.0-0. Update versions, URLs, and digests together.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from .errors import ValidationError
+
+
+CATALOG = (
+    {"id": "git", "name": "Git", "description": "Version control for your projects.",
+     "version_label": "Ubuntu package", "versions": [], "source": "Ubuntu"},
+    {"id": "nodejs", "name": "Node.js", "description": "JavaScript runtime with npm and npx.",
+     "version_label": "24 LTS · 24.21.0", "versions": [], "source": "nodejs.org"},
+    {"id": "python", "name": "Python", "description": "Side-by-side Python runtimes with pip and venv.",
+     "version_label": "3.11 / 3.12 / 3.13", "versions": ["3.11", "3.12", "3.13"],
+     "source": "Astral python-build-standalone"},
+    {"id": "jdk", "name": "JDK", "description": "OpenJDK development kit: java and javac.",
+     "version_label": "OpenJDK 21 LTS", "versions": [], "source": "Ubuntu"},
+    {"id": "wget", "name": "wget", "description": "Download files from the command line.",
+     "version_label": "Ubuntu package", "versions": [], "source": "Ubuntu"},
+    {"id": "curl", "name": "curl", "description": "Transfer data and call HTTP APIs.",
+     "version_label": "Ubuntu package", "versions": [], "source": "Ubuntu"},
+    {"id": "htop", "name": "htop", "description": "Interactive process and resource monitor.",
+     "version_label": "Ubuntu package", "versions": [], "source": "Ubuntu"},
+    {"id": "php", "name": "PHP", "description": "PHP command-line runtime.",
+     "version_label": "8.5 CLI", "versions": [], "source": "Ondřej Surý PHP PPA"},
+    {"id": "conda", "name": "Conda", "description": "Miniforge with the conda-forge package channel.",
+     "version_label": "Miniforge 26.5.3-0", "versions": [], "source": "conda-forge"},
+    {"id": "micromamba", "name": "micromamba", "description": "Lightweight Conda-compatible environment manager.",
+     "version_label": "2.9.0", "versions": [], "source": "mamba-org"},
+)
+TOOL_IDS = frozenset(tool["id"] for tool in CATALOG)
+MAX_SELECTIONS = 12
+
+
+@dataclass(frozen=True, slots=True)
+class ToolSelection:
+    id: str
+    version: str | None = None
+
+    def __post_init__(self) -> None:
+        if not isinstance(self.id, str) or self.id not in TOOL_IDS:
+            raise ValidationError("unknown tool; choose a tool from GET /tools")
+        if self.id == "python":
+            if self.version not in ("3.11", "3.12", "3.13"):
+                raise ValidationError("Python requires version 3.11, 3.12, or 3.13")
+        elif self.version is not None:
+            raise ValidationError("only Python accepts a version; other tools use the catalog default")
+
+    @property
+    def key(self) -> str:
+        return f"{self.id}-{self.version}" if self.version else self.id
+
+    def to_dict(self) -> dict:
+        return {"id": self.id, "version": self.version}
+
+
+def normalize_tools(values: object, *, allow_empty: bool = False) -> tuple[ToolSelection, ...]:
+    if not isinstance(values, (list, tuple)) or len(values) > MAX_SELECTIONS:
+        raise ValidationError(f"tools must be a list of at most {MAX_SELECTIONS} catalog selections")
+    result = {}
+    for value in values:
+        if isinstance(value, ToolSelection):
+            selection = ToolSelection(value.id, value.version)
+        elif isinstance(value, dict) and "id" in value and not set(value) - {"id", "version"}:
+            selection = ToolSelection(value["id"], value.get("version"))
+        else:
+            raise ValidationError("each tool selection requires an id and optional Python version")
+        result[selection.key] = selection
+    if not result and not allow_empty:
+        raise ValidationError("select at least one tool to install")
+    # Stable ordering also makes duplicate requests independent of checkbox order.
+    order = {tool["id"]: index for index, tool in enumerate(CATALOG)}
+    return tuple(sorted(result.values(), key=lambda item: (order[item.id], item.version or "")))
+
+
+def all_selections() -> tuple[ToolSelection, ...]:
+    return tuple(ToolSelection(tool["id"], version) for tool in CATALOG for version in tool["versions"] or [None])
+
+
+_PYTHON_RELEASE = "20260924"
+_PYTHON = {
+    "3.11": ("3.11.16", {
+        "x86_64": "68c6739376b65258dee5058ccf6777232fe38d31a578965ae8bda327ec7da3a8",
+        "aarch64": "763dae20c6abe982de86ceb7829cc050c5bda79a9824846d106cc1c86d57701b",
+    }),
+    "3.12": ("3.12.14", {
+        "x86_64": "269b2c99e4db15b242bf01832f4fea1e8f1a664f273cff519393f296e9820b41",
+        "aarch64": "c8499b61252c433280f134df954464d19811527b31cb920c35fc6967c1222e35",
+    }),
+    "3.13": ("3.13.15", {
+        "x86_64": "d0b640eed27fbdd6f5f2bd33444aee53df2c8863f8b2a96f4094717411e3de9c",
+        "aarch64": "5ad58156cbec94e5643c13caa792e92df72a23f33d3a6425d4cbff5c4b7a040c",
+    }),
+}
+_NODE_SHA256 = {
+    "x86_64": "fd8e59d5a511510f6a298afb548f18c7d2b1be404d8b4a27d94fbe49f56cb2d6",
+    "aarch64": "6ad1325edbdb5649c379b75a237147a666c95d4f9ae8d340fef2d1575d289ad2",
+}
+_CONDA_SHA256 = {
+    "x86_64": "14db468222ad564658656f769506056209b6dc375f5e7dfd31eb5ebbf08fa529",
+    "aarch64": "0391e42075a7632e9665d6e728387ee6b905f6c3e704d3513e1c1133d0d69b89",
+}
+_MICROMAMBA_SHA256 = {
+    "x86_64": "366cd9cd8be14df1ab8ed50352a82111082a36686b2d389fdb79a92c3fafb3e3",
+    "aarch64": "9f93b974adcb4d166996af969b6cd371287d1a3e52733704727884d9b74cb7a7",
+}
+
+
+def recipe(selection: ToolSelection, architecture: str) -> dict:
+    """Return a JSON-serializable recipe; no request text becomes a shell program."""
+    if architecture not in {"x86_64", "aarch64"}:
+        raise ValidationError("tool installation supports x86_64 and ARM64 guests")
+    tool = selection.id
+    result = {**selection.to_dict(), "key": selection.key, "kind": "apt", "packages": [tool],
+              "checks": [[tool, "--version"]], "match": r"(\d+(?:\.\d+)+)", "prefix_match": "",
+              "links": {}, "usage": f"{tool} --version"}
+    if tool == "nodejs":
+        arch = "x64" if architecture == "x86_64" else "arm64"
+        result.update(kind="archive", release="nodejs-24.21.0", packages=["ca-certificates", "libstdc++6"],
+                      url=f"https://nodejs.org/dist/v24.21.0/node-v24.21.0-linux-{arch}.tar.xz",
+                      sha256=_NODE_SHA256[architecture], archive_root=f"node-v24.21.0-linux-{arch}",
+                      checks=[["node", "--version"], ["npm", "--version"], ["npx", "--version"]], prefix_match="24.",
+                      links={name: f"bin/{name}" for name in ("node", "npm", "npx")}, usage="node --version; npm --version")
+    elif tool == "python":
+        version, hashes = _PYTHON[selection.version]
+        binary = f"python{selection.version}"
+        result.update(kind="archive", release=f"python-{version}", packages=["ca-certificates", "libcrypt1"],
+                      url=(f"https://github.com/astral-sh/python-build-standalone/releases/download/{_PYTHON_RELEASE}/"
+                           f"cpython-{version}+{_PYTHON_RELEASE}-{architecture}-unknown-linux-gnu-install_only_stripped.tar.gz"),
+                      sha256=hashes[architecture], archive_root="python", prefix_match=selection.version + ".",
+                      checks=[[binary, "-c", "import sys, ssl, sqlite3, venv, ensurepip; print(sys.version.split()[0])"],
+                              [binary, "-m", "pip", "--version"]], links={binary: f"bin/{binary}"},
+                      usage=f"{binary} -m venv .venv; . .venv/bin/activate")
+    elif tool == "jdk":
+        arch = "amd64" if architecture == "x86_64" else "arm64"
+        home = f"/usr/lib/jvm/java-21-openjdk-{arch}"
+        result.update(packages=["openjdk-21-jdk-headless"], checks=[["javac", "-version"], ["java", "-version"]],
+                      prefix_match="21.", java_home=home, links={name: f"{home}/bin/{name}" for name in ("java", "javac")},
+                      usage="java -version; javac -version")
+    elif tool == "php":
+        result.update(kind="php", packages=["php8.5-cli"], prefix_match="8.5.",
+                      links={"php": "/usr/bin/php8.5"}, usage="php --version")
+    elif tool == "conda":
+        result.update(kind="conda", release="miniforge-26.5.3-0", packages=["ca-certificates", "bzip2"],
+                      url=("https://github.com/conda-forge/miniforge/releases/download/26.5.3-0/"
+                           f"Miniforge3-26.5.3-0-Linux-{architecture}.sh"), sha256=_CONDA_SHA256[architecture],
+                      links={"conda": "bin/conda", "mamba": "bin/mamba"},
+                      usage='eval "$(conda shell.posix hook)"; conda create -n myenv python=3.13; conda activate myenv')
+    elif tool == "micromamba":
+        arch = "64" if architecture == "x86_64" else "aarch64"
+        result.update(kind="binary", release="micromamba-2.9.0", packages=["ca-certificates"],
+                      url=f"https://github.com/mamba-org/micromamba-releases/releases/download/2.9.0-0/micromamba-linux-{arch}",
+                      sha256=_MICROMAMBA_SHA256[architecture], links={"micromamba": "micromamba"},
+                      usage='eval "$(micromamba shell hook --shell posix)"; micromamba create -n myenv -c conda-forge python=3.13; micromamba activate myenv')
+    return result

uvm/tool_store.py

@@ -0,0 +1,197 @@
+"""Atomic per-VM tool job records, serialized with lifecycle filesystem changes."""
+
+from __future__ import annotations
+
+import copy
+import json
+import os
+import re
+import stat
+import tempfile
+import time
+import uuid
+from contextlib import contextmanager
+
+from .config import Settings
+from .errors import ConflictError, NotFoundError, StateError
+from .state import StateStore
+from .tool_catalog import normalize_tools
+
+
+ACTIVE_JOB_STATUSES = frozenset({"queued", "starting", "waiting", "installing"})
+MAX_RETAINED_JOBS = 10
+MAX_DOCUMENT_BYTES = 16 * 1024 * 1024
+
+
+class ToolJobStore:
+    """Only short transactions hold operations.lock; guest execution never does."""
+
+    def __init__(self, settings: Settings, state_store: StateStore, *, clock=time.time) -> None:
+        self.settings = settings
+        self.state = state_store
+        self.clock = clock
+
+    def _path(self, vm_id):
+        vm = self.state.get(vm_id)  # Exact ID only, including while creating a VM.
+        if not re.fullmatch(r"[A-Za-z0-9][A-Za-z0-9_.-]{0,80}", vm.id):
+            raise StateError("invalid VM ID in tool history")
+        directory = self.settings.vm_dir(vm.id)
+        try:
+            metadata = directory.lstat()
+        except FileNotFoundError as error:
+            raise NotFoundError("VM runtime directory is no longer available") from error
+        if (not stat.S_ISDIR(metadata.st_mode) or metadata.st_uid != os.geteuid()
+                or metadata.st_mode & 0o077 or not vm.runtime_owned):
+            raise StateError("tool history requires a private, owned VM runtime directory")
+        return directory / "tools.json"
+
+    @staticmethod
+    def _empty():
+        return {"schema_version": 1, "jobs": [], "inventory": [], "checked_at": None,
+                "agent_ready": False, "guest_busy": False, "supported": None, "error": None}
+
+    def _read(self, vm_id):
+        path = self._path(vm_id)
+        try:
+            descriptor = os.open(path, os.O_RDONLY | os.O_NOFOLLOW)
+        except FileNotFoundError:
+            return self._empty()
+        except OSError as error:
+            raise StateError("could not open private tool installation history") from error
+        try:
+            with os.fdopen(descriptor, "rb") as stream:
+                metadata = os.fstat(stream.fileno())
+                if (not stat.S_ISREG(metadata.st_mode) or metadata.st_nlink != 1
+                        or metadata.st_uid != os.geteuid() or metadata.st_mode & 0o077):
+                    raise StateError("unsafe tool history file")
+                raw = stream.read(MAX_DOCUMENT_BYTES + 1)
+                if len(raw) > MAX_DOCUMENT_BYTES:
+                    raise StateError("tool history exceeds its size limit")
+            data = json.loads(raw)
+            if (not isinstance(data, dict) or data.get("schema_version") != 1
+                    or not isinstance(data.get("jobs"), list) or not isinstance(data.get("inventory"), list)
+                    or len(data["jobs"]) > MAX_RETAINED_JOBS):
+                raise StateError("invalid tool history document")
+            for job in data["jobs"]:
+                if (not isinstance(job, dict) or job.get("vm_id") != vm_id
+                        or not re.fullmatch(r"tools-[0-9a-f]{32}", str(job.get("id", "")))
+                        or not isinstance(job.get("items"), list) or len(job["items"]) > 12):
+                    raise StateError("invalid tool installation job")
+            return data
+        except (OSError, ValueError) as error:
+            raise StateError("could not read tool installation history") from error
+
+    def _write(self, vm_id, data):
+        path = self._path(vm_id)  # Never recreate a deleted VM directory.
+        descriptor, name = tempfile.mkstemp(prefix=".tools-", suffix=".json", dir=path.parent)
+        try:
+            with os.fdopen(descriptor, "w", encoding="utf-8") as stream:
+                os.fchmod(stream.fileno(), 0o600)
+                json.dump(data, stream, ensure_ascii=True, allow_nan=False)
+                stream.write("\n")
+                stream.flush()
+                os.fsync(stream.fileno())
+            os.replace(name, path)
+            self.state._fsync_parent(path)
+        except OSError as error:
+            raise StateError("could not save tool installation history") from error
+        finally:
+            try:
+                os.unlink(name)
+            except FileNotFoundError:
+                pass
+
+    @contextmanager
+    def transaction(self, vm_id):
+        with self.state.operation_lock():
+            data = self._read(vm_id)
+            yield data
+            self._write(vm_id, data)
+
+    def read(self, vm_id):
+        with self.state.operation_lock():
+            return self._read(vm_id)
+
+    def enqueue_locked(self, vm_id, selections, *, start_if_stopped=False, restart=False,
+                       request_id=None, source="user"):
+        """Called by lifecycle create/submit with operations.lock already held."""
+        selections = normalize_tools(selections)
+        data = self._read(vm_id)
+        wanted = [item.to_dict() for item in selections]
+        for previous in reversed(data["jobs"]):
+            same = (previous["selections"] == wanted and previous["start_if_stopped"] == start_if_stopped
+                    and previous.get("restart", False) == restart)
+            if request_id is not None and previous.get("request_id") == request_id:
+                if not same:
+                    raise ConflictError("this installation request ID was already used for different tools")
+                return copy.deepcopy(previous)
+            if previous["status"] in ACTIVE_JOB_STATUSES:
+                if same:
+                    return copy.deepcopy(previous)
+                raise ConflictError("a tool installation is already queued or running for this VM")
+        now = self.clock()
+        job = {"id": "tools-" + uuid.uuid4().hex, "vm_id": vm_id, "status": "queued", "source": source,
+               "selections": wanted, "start_if_stopped": start_if_stopped, "restart": restart,
+               "request_id": request_id, "created_at": now, "updated_at": now, "completed_at": None,
+               "error": None, "outcome_unknown": False,
+               "items": [{**item.to_dict(), "key": item.key, "status": "queued", "detected_version": None,
+                          "error": None, "output": "", "truncated": False} for item in selections]}
+        data["jobs"] = [*data["jobs"], job][-MAX_RETAINED_JOBS:]
+        self._write(vm_id, data)
+        return copy.deepcopy(job)
+
+    def get_job(self, vm_id, job_id):
+        for job in self.read(vm_id)["jobs"]:
+            if job["id"] == job_id:
+                return job
+        raise NotFoundError("tool installation job not found for this VM")
+
+    def update_job(self, vm_id, job_id, update):
+        with self.transaction(vm_id) as data:
+            for job in data["jobs"]:
+                if job["id"] == job_id:
+                    update(job, data)
+                    job["updated_at"] = self.clock()
+                    return copy.deepcopy(job)
+        raise NotFoundError("tool installation job not found for this VM")
+
+    def claim(self):
+        with self.state.operation_lock():
+            for vm in self.state.load().vms.values():
+                # A failed/interrupted create may not own a runtime directory.
+                if not vm.runtime_owned or not self.settings.vm_dir(vm.id).is_dir():
+                    continue
+                data = self._read(vm.id)
+                for job in data["jobs"]:
+                    if job["status"] == "queued":
+                        job["status"] = "starting" if vm.status == "stopped" or job.get("restart") else "waiting"
+                        job["updated_at"] = self.clock()
+                        self._write(vm.id, data)
+                        return copy.deepcopy(job)
+        return None
+
+    @staticmethod
+    def interrupt(job, message, now, *, outcome_unknown=True):
+        job.update(status="interrupted", error=message, outcome_unknown=outcome_unknown,
+                   updated_at=now, completed_at=now)
+        for item in job["items"]:
+            if item["status"] in {"installing", "queued"}:
+                item["status"] = "interrupted" if item["status"] == "installing" else "skipped"
+                item["error"] = message
+
+    def recover(self):
+        """The worker owner lock proves that previous in-flight host workers exited."""
+        with self.state.operation_lock():
+            for vm in self.state.load().vms.values():
+                if not vm.runtime_owned or not self.settings.vm_dir(vm.id).is_dir():
+                    continue
+                data = self._read(vm.id)
+                changed = False
+                for job in data["jobs"]:
+                    if job["status"] in ACTIVE_JOB_STATUSES - {"queued"}:
+                        self.interrupt(job, "Management server restarted. Refresh tool status before retrying; "
+                                       "the last installation may have completed inside the guest.", self.clock())
+                        changed = True
+                if changed:
+                    data.update(agent_ready=False, guest_busy=False)
+                    self._write(vm.id, data)

uvm/tools.py

@@ -0,0 +1,338 @@
+"""Bounded background installation workers using the existing guest exec channel."""
+
+from __future__ import annotations
+
+import fcntl
+import json
+import logging
+import os
+import re
+import shlex
+import stat
+import threading
+import time
+from functools import lru_cache
+from pathlib import Path
+
+from . import guest_tools
+from .errors import ConflictError, GuestAgentUnavailableError, GuestExecutionError, NotFoundError, UvmError, ValidationError
+from .guest_exec import validate_exec
+from .system import require_root
+from .tool_catalog import all_selections, normalize_tools, recipe
+from .tool_store import ToolJobStore
+
+
+@lru_cache(maxsize=1)
+def _guest_source():
+    return Path(guest_tools.__file__).read_text(encoding="utf-8")
+
+
+def guest_command(payload: dict) -> str:
+    command = shlex.join(["/usr/bin/python3", "-I", "-c", _guest_source(), json.dumps(payload, ensure_ascii=True)])
+    validate_exec(command, True, 60)
+    return command
+
+
+def _identity(vm):
+    return vm.pid, vm.process_start_time, vm.boot_id
+
+
+def _text(value, limit=16384):
+    return str(value or "").encode("utf-8", errors="replace")[-limit:].decode("utf-8", errors="replace")
+
+
+class ToolService:
+    def __init__(self, settings, state_store, lifecycle) -> None:
+        self.settings = settings
+        self.state = state_store
+        self.lifecycle = lifecycle
+        self.store = ToolJobStore(settings, state_store)
+        self._mutex = threading.RLock()
+        self._wake = threading.Condition()
+        self._stop = threading.Event()
+        self._threads = []
+        self._sessions = {}
+        self._owner = None
+        self._workers_left = 0
+
+    def start(self):
+        with self._mutex:
+            if self._stop.is_set():
+                raise ConflictError("the tool installer is shutting down")
+            if self._owner is not None:
+                return
+            with self.state.operation_lock():
+                descriptor = os.open(self.settings.base / "tool-worker.lock", os.O_RDWR | os.O_CREAT | os.O_NOFOLLOW, 0o600)
+                try:
+                    metadata = os.fstat(descriptor)
+                    if not stat.S_ISREG(metadata.st_mode) or metadata.st_nlink != 1 or metadata.st_uid != os.geteuid():
+                        raise UvmError("unsafe tool worker ownership file")
+                    os.fchmod(descriptor, 0o600)
+                    fcntl.flock(descriptor, fcntl.LOCK_EX | fcntl.LOCK_NB)
+                except BaseException as error:
+                    os.close(descriptor)
+                    if isinstance(error, BlockingIOError):
+                        raise ConflictError("another management server is already processing tool installations for this registry") from error
+                    raise
+            self._owner = descriptor
+            try:
+                self.store.recover()
+                for index in range(self.settings.tool_workers):
+                    thread = threading.Thread(target=self._worker, name=f"uvm-tools-{index}", daemon=True)
+                    self._workers_left += 1
+                    try:
+                        thread.start()
+                    except BaseException:
+                        self._workers_left -= 1
+                        raise
+                    self._threads.append(thread)
+            except BaseException:
+                self._stop.set()
+                self.notify()
+                if not self._workers_left:
+                    self._release_owner()
+                raise
+
+    def _release_owner(self):
+        if self._owner is not None:
+            os.close(self._owner)
+            self._owner = None
+
+    def notify(self):
+        with self._wake:
+            self._wake.notify_all()
+
+    def shutdown(self):
+        self._stop.set()
+        self.notify()
+        with self._mutex:
+            sessions = list(self._sessions.values())
+        for session in sessions:
+            session.close()
+        deadline = time.monotonic() + 8
+        for thread in self._threads:
+            thread.join(max(0, deadline - time.monotonic()))
+        # A worker finishing a host start keeps ownership until that start ends.
+        # Its guest command is never submitted after shutdown was requested.
+        with self._mutex:
+            if not self._workers_left:
+                self._release_owner()
+
+    def submit(self, vm_id, selections, *, start_if_stopped=False, restart=False, request_id=None):
+        require_root()
+        selections = normalize_tools(selections)
+        if type(start_if_stopped) is not bool or type(restart) is not bool:
+            raise ValidationError("start_if_stopped and restart must be booleans")
+        if request_id is not None and (not isinstance(request_id, str) or not re.fullmatch(r"[A-Za-z0-9_-]{1,80}", request_id)):
+            raise ValidationError("request_id must contain 1–80 letters, numbers, underscores, or hyphens")
+        self.start()
+        job = self.lifecycle.queue_tools(vm_id, selections, start_if_stopped=start_if_stopped,
+                                         restart=restart, request_id=request_id)
+        self.notify()
+        return job
+
+    def _execute(self, vm, payload, timeout):
+        if self._stop.is_set():
+            raise InterruptedError("management server is shutting down")
+        thread_id = threading.get_ident()
+
+        def connected(session):
+            with self._mutex:
+                self._sessions[thread_id] = session
+                if self._stop.is_set():
+                    session.close()
+
+        try:
+            result = self.lifecycle.execute(vm.id, guest_command(payload), sudo=True, timeout_seconds=timeout,
+                                            expected_identity=_identity(vm), on_connect=connected)
+        finally:
+            with self._mutex:
+                self._sessions.pop(thread_id, None)
+        if result.timed_out:
+            return {"ok": False, "error": f"installation/verification timed out after {timeout:g} seconds; retry to finish it",
+                    "output": _text(result.stderr), "truncated": result.truncated}
+        try:
+            value = json.loads(result.stdout)
+            if result.exit_code != 0 or result.truncated or not isinstance(value, dict) or type(value.get("ok")) is not bool:
+                raise ValueError("invalid result")
+            return value
+        except (ValueError, TypeError) as error:
+            raise GuestExecutionError("guest returned an incomplete tool result; refresh status before retrying",
+                                      outcome_unknown=payload["action"] == "install") from error
+
+    def _payload(self, vm, action, selections, *, boot_id=None):
+        return {"action": action, "username": vm.username, "architecture": self.settings.architecture,
+                "boot_id": boot_id, "tools": [recipe(selection, self.settings.architecture) for selection in selections]}
+
+    @staticmethod
+    def _inventory(result, selections):
+        expected = {item.key: item for item in selections}
+        records = result.get("tools")
+        if not isinstance(records, list) or len(records) != len(expected):
+            raise GuestExecutionError("invalid guest tool inventory")
+        inventory = []
+        for value in records:
+            if (not isinstance(value, dict) or value.get("key") not in expected
+                    or type(value.get("installed")) is not bool):
+                raise GuestExecutionError("invalid guest tool inventory")
+            selection = expected.pop(value["key"])
+            inventory.append({**selection.to_dict(), "key": selection.key, "installed": value["installed"],
+                              "detected_version": _text(value.get("detected_version"), 120) or None,
+                              "error": _text(value.get("error"), 512) or None,
+                              "usage": _text(value.get("usage"), 512)})
+        return inventory
+
+    def snapshot(self, vm_id, *, refresh=False):
+        require_root()
+        vm = self.state.get(vm_id)
+        data = self.store.read(vm_id)
+        running = self.lifecycle.inspect(vm.id).observed_status == "running"
+        if refresh and running:
+            supported = data["supported"]
+            try:
+                result = self._execute(vm, self._payload(vm, "probe", all_selections()), 90)
+                if type(result.get("supported")) is bool:
+                    supported = result["supported"]
+                if not result["ok"]:
+                    raise UvmError(_text(result.get("error"), 2048))
+                inventory = self._inventory(result, all_selections())
+                with self.store.transaction(vm_id) as current:
+                    current.update(inventory=inventory, checked_at=time.time(), error=None, agent_ready=True,
+                                   guest_busy=result.get("busy") is True, supported=result.get("supported") is True)
+                data = self.store.read(vm_id)
+            except UvmError as error:
+                with self.store.transaction(vm_id) as current:
+                    current.update(error=_text(error, 2048), agent_ready=not isinstance(error, GuestAgentUnavailableError),
+                                   supported=supported)
+                data = self.store.read(vm_id)
+        if not running:
+            data.update(agent_ready=False, guest_busy=False)
+        return {"vm_id": vm_id, **{key: data[key] for key in (
+            "inventory", "checked_at", "agent_ready", "guest_busy", "supported", "error")},
+            "stale": not running or data["checked_at"] is None or time.time() - data["checked_at"] > 60,
+            "latest_job": data["jobs"][-1] if data["jobs"] else None,
+            "jobs": [{key: job[key] for key in ("id", "status", "created_at", "completed_at")}
+                     for job in reversed(data["jobs"])]}
+
+    def _worker(self):
+        try:
+            while not self._stop.is_set():
+                try:
+                    job = self.store.claim()
+                    if job is not None:
+                        self._run_job(job)
+                        continue
+                except (UvmError, OSError):
+                    logging.getLogger(__name__).exception("Could not process a tool installation job")
+                with self._wake:
+                    self._wake.wait(timeout=1)
+        finally:
+            with self._mutex:
+                self._workers_left -= 1
+                if self._workers_left == 0 and self._stop.is_set():
+                    self._release_owner()
+
+    def _update(self, job, update):
+        return self.store.update_job(job["vm_id"], job["id"], update)
+
+    def _wait_ready(self, vm, selections):
+        deadline = time.monotonic() + self.settings.tool_ready_timeout_s
+        while not self._stop.is_set():
+            try:
+                result = self._execute(vm, self._payload(vm, "probe", selections), 90)
+                if not result["ok"]:
+                    raise UvmError(_text(result.get("error"), 2048))
+                if result.get("busy"):
+                    raise ConflictError("an earlier installation is still running inside this VM; wait and refresh before retrying")
+                if not isinstance(result.get("boot_id"), str) or not result["boot_id"]:
+                    raise GuestExecutionError("guest did not report its boot identity")
+                return result
+            except GuestAgentUnavailableError:
+                if time.monotonic() >= deadline:
+                    raise GuestAgentUnavailableError("Guest agent did not become ready. Restart & retry to apply agent setup, "
+                                                     "then check the VM log if startup still fails.")
+                self._stop.wait(min(2, max(0, deadline - time.monotonic())))
+        raise InterruptedError("management server is shutting down")
+
+    def _run_job(self, job):
+        submitted = False
+        try:
+            if self._stop.is_set():
+                raise InterruptedError("management server is shutting down")
+            selections = normalize_tools(job["selections"])
+            vm = self.lifecycle.prepare_tool_install(job["vm_id"], start_if_stopped=job["start_if_stopped"],
+                                                     restart=job.get("restart", False))
+            self._update(job, lambda current, data: current.update(status="waiting"))
+            info = self._wait_ready(vm, selections)
+            initial = self._inventory(info, selections)
+
+            def ready(current, data):
+                observed = {item["key"]: item for item in data["inventory"]}
+                observed.update({item["key"]: item for item in initial})
+                data.update(inventory=list(observed.values()), agent_ready=True,
+                            guest_busy=False, supported=True, error=None)
+            self._update(job, ready)
+            for index, selection in enumerate(selections):
+                if self._stop.is_set():
+                    raise InterruptedError("management server is shutting down")
+
+                def begin(current, data):
+                    current["status"] = "installing"
+                    current["items"][index]["status"] = "installing"
+                self._update(job, begin)
+                submitted = True
+                result = self._execute(vm, self._payload(vm, "install", [selection], boot_id=info["boot_id"]),
+                                       self.settings.tool_install_timeout_s)
+                observed = None
+                if result["ok"]:
+                    observed, = self._inventory({"tools": [result.get("tool")]}, [selection])
+                    if not observed["installed"] or result.get("status") not in {"installed", "already_installed"}:
+                        raise GuestExecutionError("guest installation did not verify the selected tool", outcome_unknown=True)
+
+                def finish_item(current, data):
+                    current["items"][index].update(
+                        status=result["status"] if result["ok"] else "failed",
+                        detected_version=observed["detected_version"] if observed else None,
+                        error=None if result["ok"] else _text(result.get("error"), 2048),
+                        output=_text(result.get("output")), truncated=result.get("truncated") is True)
+                    inventory = {item["key"]: item for item in data["inventory"]}
+                    if observed:
+                        inventory[selection.key] = observed
+                    elif selection.key in inventory:
+                        inventory[selection.key].update(installed=False, error="Refresh to verify after the failed installation.")
+                    data.update(inventory=list(inventory.values()))
+                self._update(job, finish_item)
+                submitted = False
+
+            def finish(current, data):
+                failures = sum(item["status"] == "failed" for item in current["items"])
+                current.update(status="failed" if failures else "succeeded", completed_at=time.time(),
+                               error=f"{failures} tool installation(s) failed; see the per-tool output and retry." if failures else None)
+            self._update(job, finish)
+        except NotFoundError:
+            return  # Deletion owns removal of the whole runtime, including history.
+        except (InterruptedError, GuestExecutionError, ConflictError) as error:
+            try:
+                self._update(job, lambda current, data: self.store.interrupt(
+                    current, _text(error, 2048), time.time(),
+                    outcome_unknown=getattr(error, "outcome_unknown", submitted)))
+            except NotFoundError:
+                pass
+        except (UvmError, OSError, ValueError) as error:
+            def failed(current, data):
+                self.store.interrupt(current, _text(error, 2048), time.time(), outcome_unknown=submitted)
+                current["status"] = "failed" if not submitted else "interrupted"
+                if isinstance(error, GuestAgentUnavailableError):
+                    data.update(agent_ready=False, error=_text(error, 2048))
+            try:
+                self._update(job, failed)
+            except NotFoundError:
+                pass
+        except Exception:
+            logging.getLogger(__name__).exception("Unexpected tool installation failure")
+            try:
+                self._update(job, lambda current, data: self.store.interrupt(
+                    current, "Unexpected installer failure; check the management server log and refresh before retrying.",
+                    time.time(), outcome_unknown=submitted))
+            except NotFoundError:
+                pass