Skip to content

v0.4.3

Choose a tag to compare

@github-actions github-actions released this 21 Jul 19:06

Upgrade notes

  • agentos cluster down now returns the retryable exit class (3) instead of 1 when teardown
    fails for a connectivity reason, and it sweeps the namespaces it created even when
    helm uninstall fails, so compute is no longer orphaned behind an unreachable API server.
    Scripts branching on exit code 1 need updating.
  • agentos local message and agentos cluster message now wait for an approval to resolve
    instead of returning as soon as the turn is parked. The wait is bounded by the existing
    --timeout-secs. Scripted callers that relied on the immediate return will now block.
  • agentos cluster message no longer falls back to the dev credentials. On a default
    cluster up install it reads the release's own API key and Valkey password from the chart
    Secret; an explicit flag or env var still wins. This fixes the 401 that made the documented
    cluster round trip fail on a strong-secrets install.
  • agentos skill down with no .agentos/runner.json now exits 0 instead of 1.

What's Changed

  • Warn on mistaken approval-gate names; drop internal jargon from the default booting placeholder by @alexrao2000 in #719
  • Give the sandbox docker-run network error an actionable remediation hint by @alexrao2000 in #720
  • Add an operator command to force a fresh sandbox for a stuck thread by @alexrao2000 in #721
  • Nudge new agents toward the greeting behavior pack in the scaffolded AGENTS.md by @alexrao2000 in #722
  • Add agentos local rebuild to fix single-service credential loss by @alexrao2000 in #723
  • Log claim latency separately from turn duration for diagnosability by @alexrao2000 in #724
  • Render ConfirmIntent/ChoiceIntent's prompt in the Slack approval card by @alexrao2000 in #725
  • Add a JavaScript dependency audit alongside the Rust/Python ones by @alexrao2000 in #726
  • Wire AGENTOS_FALSE_COMPLETION_CHECK through the worker to every sandbox boot by @alexrao2000 in #730
  • Gate the tag-triggered release pipeline on reviewed-main ancestry and green checks by @alexrao2000 in #727
  • Retire e2e.sh's cluster-verb-against-compose deploy leg by @alexrao2000 in #728
  • Add an env-var image/tag override for compose.dev.yaml's api, migrate, and worker services by @alexrao2000 in #729
  • Bound the thread-reset interrupt so a wedged sandbox still releases by @TheConnMan in #744
  • Correct the approval-gate built-in tool recognition set by @TheConnMan in #741
  • Fix the release authorization gate so a valid tag can publish by @TheConnMan in #740
  • Record ADR-0059: the sandbox is a bounded resource envelope by @TheConnMan in #754
  • Make the install snippets platform-aware and lead onboarding with the offline path by @TheConnMan in #753
  • Exclude venvs, caches and ignore-file entries from plugin bundles by @TheConnMan in #762
  • Record a package-shaped redesign of the harness-adapter seam by @TheConnMan in #763
  • Make a leftover runner container recoverable from the CLI by @TheConnMan in #764
  • Keep the message reply surface alive until the approval resumes by @TheConnMan in #772
  • Fail-forward cluster down teardown on an unreachable API server by @TheConnMan in #769
  • Add list-agents and deploy-local, plus a bundle picker for Deploy to Slack by @alexrao2000 in #731
  • Guard every publishing job the ancestor-graph skip reached by @TheConnMan in #788
  • Discover the release credentials for cluster message by @TheConnMan in #789

Full Changelog: v0.4.2...v0.4.3