Skip to content

v0.13.1 · Reliable agent wake boundaries

Choose a tag to compare

View all tags
@Codename-11 Codename-11 released this 14 Jul 18:34
· 85 commits to main since this release
v0.13.1
d0426cd
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Reliable agent wake boundaries

Forge now treats activity, notifications, and executable agent work as separate concerns.

Changed

  • Waiting agents wake only for actionable human replies, explicit safe mentions, assignment/restart, or a real escalation into HIGH/URGENT.
  • RUNS agents resume through a fresh provider turn carrying the canonical operator input; webhook and completions agents reactivate directly.
  • Labels, ordinary status changes, non-escalating priority changes, stall/SLA markers, coach output, blocking markers, and other system activity remain visible without invoking an LLM.

Fixed

  • Agent-authored terminal/stall replies cannot wake their author or create replacement runs.
  • Nudges dispatch once instead of producing two wake deliveries.
  • AGENT_RUN_BLOCKED remains an operator-facing alert without bouncing back to the parked agent.
  • Completed, stalled, and abandoned runs atomically create exactly one durable issue reply, even with concurrent pollers.

Validation

  • Hosted lint, typecheck, and unit suite passed.
  • Hosted Playwright shards passed.
  • Local full suite: 1,236 passed, 1 intentional live-connector skip.
  • Local production build and all 38 serial Playwright journeys passed.

Merged in #26.

Assets 2
Loading