Skip to content

Recovered-only timeout alerts can start an alert-watcher investigation session #944

Open

Description

@alex-mextner

What

Found by the single-seat Opus review of #941 (work of #935), finding listed in #935 (comment). Finding 4. timeout-reporter.ts queues one alert per kind and window even when every timeout was recovered by a fallback, and hcb-alert-watcher.ts launches an omp session for any pending alert. Launch only when unrecovered is above zero or a burst threshold of recovered timeouts is passed; the recovered counts stay in the alert text.

Why (motivation)

Recovered timeouts are expected noise on a loaded host; each session costs an investigation.

User impact

Every absorbed provider timeout could open an unattended session on the owner's PC.

Cost of inaction

The owner gets investigation sessions for non-problems and learns to ignore real ones.

Acceptance criteria

  • Test: an alert batch with recovered timeouts only starts no session
  • Test: unrecovered above zero, or recovered above the threshold, starts one session

Screenshots

(none)

Deploy

Deployed: no

Links

  • (none)

Skipped gates

  • duplicate: searched incident, alert, watcher, validator, timeout, receipt; no ticket covers this
  • screenshots: no UI change

Activity

  1. alex-mextner commented on Nov 10, 2026

    @alex-mextner
    OwnerAuthor

    Live example 2026-11-10 02:31Z: watcher session hcb-alert-20261110T023129Z was opened by two recovered-only provider timeout alerts (unrecovered=0, one turn; both timeouts absorbed by the Claude fallback; the run ended normal and delivered). The measured cause is tracked separately in #978.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions