fix(desktop): normalise timeout/error subagent statuses to terminal (#73728)

[RelaxJonh]

Summary

Fixes #73728. The backend emits terminal statuses including timeout and error in subagent.complete payloads, but asStatus() only recognised completed, failed, interrupted, and queued. Unrecognised values fell through to running, making timed-out subagents immortal in the active status stack.

Changes

• apps/desktop/src/store/subagents.ts: asStatus() now maps timeout/error → failed and cancelled/canceled → interrupted. Nonterminal unknown statuses still default to running for forward compatibility.
• apps/desktop/src/store/subagents.test.ts: Added regression test covering all four backend terminal status strings (timeout, error, cancelled, canceled) with verification that they are recognised as terminal and pruned by pruneFinishedSessionSubagents().

Root cause

// Before: only recognised 4 statuses, everything else → running
const asStatus = (v: unknown): SubagentStatus =>
  v === 'completed' || v === 'failed' || v === 'interrupted' || v === 'queued' ? v : 'running'

The backend's delegate_tool.py uses timeout as a terminal status, and error paths use error. Both were normalised to running, so pruneFinishedSessionSubagents() kept them forever and they rendered as active in the composer status stack.

Test plan

• Existing test suite: all pass
• New regression test: 4 backend terminal statuses → recognised → pruned

[teknium1]

Thanks — this addresses the verified current mismatch: tools/delegate_tool.py:2243-2251 emits subagent.complete with timeout or error, while current Desktop normalization in apps/desktop/src/store/subagents.ts:58-59 turns both into running.

Problems

• apps/desktop/src/store/subagents.ts:62 still maps every unknown status to running without considering the event type. An unknown terminal value on subagent.complete would therefore remain active and survive the running/queued-only prune at apps/desktop/src/store/subagents.ts:212. This is the future-terminal-status case called out in #73728.

Suggested changes

• Make normalization event-aware: map unknown values on subagent.complete to failed, while retaining running for unknown nonterminal events. Add the corresponding prune regression test.

Automated hermes-sweeper review.

[teknium1]

asStatus() cannot distinguish a progress update from subagent.complete. Please map an unknown value on a completion event to failed (and test pruning it), while keeping unknown nonterminal events as running; otherwise a future terminal backend status recreates the stuck-active-row bug.

[GottZ]

This was generated by AI during triage.

Summary

One PR addresses issue #73728. #73859 fixes the reproduced timeout/error and cancellation normalization paths, but its event-agnostic fallback does not eliminate the underlying failure mode for future unknown completion statuses.

Related pull requests

• fix(desktop): normalise timeout/error subagent statuses to terminal (#73728) #73859 best fix — (+43/-2) — best existing partial fix, keep open with a salvage path: the diff maps timeout/error to failed and cancelled/canceled to interrupted, with regression coverage confirming those rows are pruned. Consistent with the keep_open review on fix(desktop): normalise timeout/error subagent statuses to terminal (#73728) #73859, asStatus() still maps every other unknown value to running without considering subagent.complete, so a future terminal backend status could recreate the stuck active row.

Suggested consolidation

Keep #73859 open with a salvage path: retain its targeted terminal-status mappings and pruning regression coverage, then make normalization event-aware so an unknown status on subagent.complete maps to failed while an unknown nonterminal status remains running; add a pruning test for the unknown-completion case requested in the contributor review. There are no duplicate PRs to close.

Complex graph

flowchart LR
    classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
    classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d
    classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937
    classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151
    classDef best stroke-width:3px,stroke:#b45309
    classDef target stroke-width:3px,stroke:#4338ca
    I73728(["issue #73728 (open)"])
    P73859["PR #73859 (open)"]
    P73859 -->|best fix| I73728
    class I73728 open
    class P73859 open
    class P73859 best
    class P73859 target
    click I73728 "https://github.com/NousResearch/hermes-agent/issues/73728"
    click P73859 "https://github.com/NousResearch/hermes-agent/pull/73859"

Loading

Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed (state tag in the node label).

Cross-PR triage: Reviewed 1 pull request and 1 issue in this complex. Each diff was read against this issue; Assessment working set: 3 kB of PR diffs, 5 kB of issue/PR text, 3 kB of discussion (4 comments), 3 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.

[DavidMetcalfe]

Thanks for the fix — it correctly identifies the normalization gap and maps the known statuses. To help move this forward following the maintainer review (which requested event-aware handling of unknown completion statuses), I've opened #85493 building on top of your work: your commit is cherry-picked in with authorship preserved, and it adds the event-aware fail-closed fallback, timeout summary synthesis, and tests. If you'd prefer to push updates directly to this branch instead, let me know and I'm glad to defer to your PR.

[teknium1]

Landed on main via #86592 (merge b948248), which cherry-picked your normalisation commit as-is — it is on main as a351c17 with your authorship, plus the event-aware fail-closed and timeout-summary follow-ups from #85493. Thanks @RelaxJonh; closing since the branch is fully contained in main.