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
{{ message }}
Repository navigation
Timeline: per-Bot turn outcomes and recent failures #530
#357 (performance/quality board) and #471 (agent inbox: failed turns) want reliability visible, but the timeline window only counts errors per Bot; it does not say how a Bot's turns ended or list what failed.
Expected behavior
timeline/window gains outcomes (per Bot lane: completed / aborted / error / lost) and failures (the recent failed or aborted turns, each with a cleaned error code, newest first, capped at 20).
Read-only, part of the existing answer, ids and codes only (an error code is cleaned, never provider text).
Change scope
plugins/timeline/src/window.js (the two fields), a case in plugins/timeline/test/window.test.js, docs/subsystems/timeline.md. Plugin only.
Why
A Bot's reliability becomes visible where the run is already drawn (the quality-visible bet).
Triage (Birfy, 2026-10-20 20:08 Asia/Shanghai). Read-only timeline analytics: timeline/window gains outcomes (completed/aborted/error/lost per Bot) and failures (last 20 failed or aborted turns, cleaned error codes). Plugin only. HugoXK's PR #531 implements it (reviewed PASS at 03af6d7, CI green).
Design check: no conflict with the owner rules (local, read-only, ids/codes only). It overlaps #471 (agent inbox: failed turns, needs-human) and the existing utilization[].errors. The data is already derivable from turns[], and no panel view reads the new fields yet. It is part of a series (#530/#534/#536/#538) that all add fields to the same timeline/window answer. They don't overlap the merged export work (#484/#523/#524).
Proposed scope if accepted: as in #531, plus one line in the panel's utilization row (completed/aborted/lost counts and a recent-failures list), so the field has a consumer.
needs-human: the owner needs to confirm this new feature and whether HugoXK's #531 is accepted as the implementation.
Owner decision (2026-10-20 21:09 Asia/Shanghai): accepted (option a). PR #531 is accepted; the outcomes/recent-failures data must also be visible in the dashboard/panel, not only in the timeline/window API.
Implemented by PR #531 (merged as 03bfc468, 2026-10-21 01:55Z), which adds the per-Bot outcomes counts (completed / aborted / error / lost) and a capped failures list to the existing timeline/window answer.
The PR body referenced #357/#471 only, so GitHub did not auto-close this issue. Closing it by hand with that evidence; agent-ready and agent-working come off.
Activity
Birfy commentedon Oct 20, 2026
Triage (Birfy, 2026-10-20 20:08 Asia/Shanghai). Read-only timeline analytics:
timeline/windowgainsoutcomes(completed/aborted/error/lost per Bot) andfailures(last 20 failed or aborted turns, cleaned error codes). Plugin only. HugoXK's PR #531 implements it (reviewed PASS at 03af6d7, CI green).Design check: no conflict with the owner rules (local, read-only, ids/codes only). It overlaps #471 (agent inbox: failed turns,
needs-human) and the existingutilization[].errors. The data is already derivable fromturns[], and no panel view reads the new fields yet. It is part of a series (#530/#534/#536/#538) that all add fields to the sametimeline/windowanswer. They don't overlap the merged export work (#484/#523/#524).Proposed scope if accepted: as in #531, plus one line in the panel's utilization row (completed/aborted/lost counts and a recent-failures list), so the field has a consumer.
needs-human: the owner needs to confirm this new feature and whether HugoXK's #531 is accepted as the implementation.Birfy commentedon Oct 20, 2026
Owner decision (2026-10-20 21:09 Asia/Shanghai): accepted (option a). PR #531 is accepted; the outcomes/recent-failures data must also be visible in the dashboard/panel, not only in the timeline/window API.
Birfy commentedon Oct 21, 2026
Implemented by PR #531 (merged as
03bfc468, 2026-10-21 01:55Z), which adds the per-Botoutcomescounts (completed / aborted / error / lost) and a cappedfailureslist to the existingtimeline/windowanswer.The PR body referenced #357/#471 only, so GitHub did not auto-close this issue. Closing it by hand with that evidence;
agent-readyandagent-workingcome off.