fix(cron): finalize kanban worker sessions with ended_at (#76914)

[webtecnica]

Summary

Kanban worker sessions completed their tasks but left sessions.ended_at NULL in state.db (reproduced 2/2 runs). The worker exits via the quiet single-query CLI path, which never finalized the SQLite session.

Root Cause

Kanban workers are spawned as hermes -p <profile> --cli chat -q "work kanban task <id>" (hermes_cli/kanban_db.py:8964). The quiet path in cli.py (_finalize_single_query, ~L1342) only notifies session-finalize hooks and runs cleanup — it never calls agent.close(), which is what invokes session_db.end_session(session_id, "agent_close") (run_agent.py:4150-4157, gated by _end_session_on_close, default True).

The interactive run() loop closes the session in its finally block (cli.py:17718-17721), but the quiet path bypasses it entirely. Compare hermes_cli/oneshot.py:461-481 (hermes -z calls agent.close() in its own finally) and cron/scheduler.py:3829-3832 (end_session(..., "cron_complete") explicitly).

Change

cli.py — _finalize_single_query now calls agent.close() (defensively, via getattr/callable so stub agents in tests are unaffected) before releasing the active session lease. close() is idempotent in the run loop path and triggers end_session → ended_at set.

Verification

• pytest tests/cli/test_single_query_session_finalize.py tests/cli/test_chat_q_exit_clear.py — 9 passed
• pytest tests/hermes_cli/test_kanban_lifecycle_hooks.py::test_claim_fires_hook tests/hermes_cli/test_kanban_write_guard.py — 4 passed
• 7 kanban-collection failures on the full -k kanban run reproduce on unpatched upstream/main (pre-existing order-dependent flakes, not caused by this change)

Closes #76914

[teknium1]

Thanks for tracing the Kanban worker lifecycle. The premise is confirmed on current main: hermes_cli/kanban_db.py:9131-9134 launches workers through chat -q, while cli.py:1341-1347 does not call agent.close(); run_agent.py:4096-4100 is the path that ends an owned SQLite session.

Problems

• The new close() is after _run_cleanup() (cli.py:1352 in this diff). If cleanup raises, it is never reached. tests/cli/test_single_query_session_finalize.py:16-34 already exercises a cleanup exception and proves the finalizer must continue its teardown path.

Suggested changes

• Put the guarded agent.close() in a finally before _release_active_session(), with release protected by its own finally.
• Extend tests/cli/test_single_query_session_finalize.py to assert close() runs on both normal and cleanup-failure paths.

This is an automated hermes-sweeper review.

[teknium1]

_run_cleanup() can raise (see tests/cli/test_single_query_session_finalize.py:16-34), which skips this call and leaves the owned session open. Move this guarded close into a finally that runs before the lease-release finally, and add coverage for that error path.

[alt-glitch]

This was generated by AI during triage.

Related: #50881 covers the broader one-shot session-finalization path, including explicit DB finalization and WAL flushing. This PR is the focused agent.close() repair for the #76914 ended_at regression.

[GottZ]

This was generated by AI during triage.

Summary

One PR addresses #76914. #76995 targets the reported quiet-path lifecycle gap by adding agent.close(), but the call remains after cleanup and can therefore be skipped when cleanup raises.

Related pull requests

• fix(cron): finalize kanban worker sessions with ended_at (#76914) #76995 related — (+7/-0) — partial fix: the diff adds the missing guarded agent.close() call to _finalize_single_query, which finalizes the owned SQLite session after successful cleanup. The keep_open review on fix(cron): finalize kanban worker sessions with ended_at (#76914) #76995 remains applicable because _run_cleanup() runs before the new call, so a cleanup exception still leaves sessions.ended_at unset.

Suggested consolidation

Keep #76995 open with a salvage path: move the guarded agent.close() into a finally that executes before the lease-release finally, and add coverage showing that close() runs after both successful cleanup and cleanup failure. The current revision should not be merged because the contributor's keep_open review identifies a documented error path in which the reported session-finalization defect remains.

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
    I76914(["issue #76914 (open)"])
    P76995["PR #76995 (open)"]
    P76995 -->|best fix| I76914
    class I76914 open
    class P76995 open
    class P76995 best
    class P76995 target
    click I76914 "https://github.com/NousResearch/hermes-agent/issues/76914"
    click P76995 "https://github.com/NousResearch/hermes-agent/pull/76995"

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: <1 kB of PR diffs, 4 kB of issue/PR text, 1 kB of discussion (2 comments), 2 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.