chore: version packages

[github-actions]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@gemstack/ai-autopilot@0.3.0

Minor Changes

• 0823cfa: Runner: DockerRunner — the first sandboxed adapter, running agent-authored code in a container.

  DockerRunner boots each workspace as a container via the docker CLI (no npm dependency), so untrusted, agent-authored code runs isolated from the host: its own filesystem, process space, and — with preview — a published port mapped to an ephemeral host port. It satisfies the same Runner contract as LocalRunner (fs / exec / start / preview / dispose), so it drops in behind the seam unchanged: new DockerRunner({ image?, previewPort?, previewHost? }).

  Where LocalRunner runs commands unsandboxed on the host (trusted dev/CI), DockerRunner is the one to reach for when the code is untrusted. It requires a running Docker daemon and the docker CLI on PATH; the default node:20-alpine base image carries node/npm and a POSIX shell. dockerAvailable() reports whether a daemon is reachable so callers (and the test suite) can skip cleanly when it isn't.

  WebContainer and Flue remain the still-parked sandboxed adapters (ai-autopilot: sandboxed runner adapters (Docker / WebContainer / Flue) #109).

• 7ce5bae: Runner seam: long-running processes + reachable previews (boot-and-serve).

  RunnerSession.start(command) launches a long-running command (a dev server) in the background and returns a RunnerProcess handle ({ command, exit, stop() }) — unlike exec, which awaits the command to finish. preview({ waitMs }) now waits for the port to accept connections before resolving, so the URL is live on return. A start_server runner tool exposes this to agents.

  LocalRunner implements it for real: start spawns in its own process group so stop() (and dispose()) kill the whole tree; preview polls the port. FakeRunner mirrors it for tests. This is the contract every sandboxed adapter (Docker / WebContainer / Flue) must satisfy, and it's what makes "produce a running app" reachable end to end.

• 493cc11: Bootstrap: serveCheck — a production-grade check that actually boots and serves the app.

  Until now the full-fledged loop gated on a prompt verdict (loopChecklist asks the model whether the app is production-grade). serveCheck(session, { serve, install?, build?, port?, healthPath? }) gives it teeth: inside the build's runner session it installs, optionally builds, starts the dev server, previews until the port is reachable, and fetches a health path — turning any failure (install error, server exits on boot, a 5xx, unreachable) into a concrete { blockers } verdict the improve loop then addresses. It satisfies the checklist step contract, and mergeChecklists(...) unions several checks so a pass must BOTH read production-grade AND actually run: mergeChecklists(loopChecklist({ loop }), serveCheck(session, { serve: 'npm run dev' })). A runner that can't start/preview skips the check (passing, with a note) instead of blocking. Built on the Runner seam: long-running processes (start) + readiness-checked preview (boot-and-serve) #137 runner seam.

@gemstack/example-autopilot-quickstart@0.0.2

Patch Changes

• Updated dependencies [0823cfa]
• Updated dependencies [7ce5bae]
• Updated dependencies [493cc11]
  • @gemstack/ai-autopilot@0.3.0

@gemstack/example-bootstrap-quickstart@0.0.2

Patch Changes

• Updated dependencies [0823cfa]
• Updated dependencies [7ce5bae]
• Updated dependencies [493cc11]
  • @gemstack/ai-autopilot@0.3.0