You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Make direct tool activity an aggregate of all in-flight work and interpret tool completions as success, failure, or unknown so parallel and out-of-order Hermes activity remains correct.
Acceptance criteria
Multiple concurrent tool starts are tracked independently.
Completing one tool does not make the pet idle while another tool remains in flight.
Unique completions arriving out of order settle the matching work and apply their cultivation effect at most once.
Structured tool results are classified as success, failure, or unknown.
Malformed or unsupported results are unknown and neutral rather than treated as success or failure.
Activity priority is active direct tool work, then a recent tool result, then smart-mode inference, then idle; later tickets may insert higher-priority approval and subagent states.
A recent outcome remains visible for a bounded presentation interval without ending unrelated in-flight work or reapplying cultivation.
Real-HTTP integration tests and fake hook-context tests cover parallel starts, out-of-order completions, and all three outcomes.
Parent
#1
What to build
Make direct tool activity an aggregate of all in-flight work and interpret tool completions as success, failure, or unknown so parallel and out-of-order Hermes activity remains correct.
Acceptance criteria
Blocked by
#3.