fix(kanban): snap notify-sub cursor to current MAX(id) at creation

[briandevans]

What does this PR do?

add_notify_sub previously inserted kanban_notify_subs rows with last_event_id = 0 (the schema default). A subscription created on an already-active task therefore replayed every prior terminal event on the next _kanban_notifier_watcher tick — issue #29905 reports a 100+ message burst at gateway boot caused by 27 stale subs at last_event_id=0.

After this PR, add_notify_sub snaps the cursor to COALESCE(MAX(id) FROM task_events WHERE task_id = ?, 0) inside the same write_txn as the INSERT, so a new sub starts caught up. Only events generated after subscription fire notifications. The snap is gated on cur.rowcount > 0, so the existing INSERT OR IGNORE semantics are preserved: a re-subscription against an existing (task, platform, chat, thread) tuple is still a no-op and never rewinds the advancing cursor.

Scope is limited to the root cause (the missing snap on creation). Existing subs already sitting at last_event_id=0 are out of scope here — the reporter's workaround SQL handles those, and a separate on-connect migration would be a larger, riskier change to bundle.

Audited siblings: add_notify_sub is the only production path that inserts into kanban_notify_subs (verified via rg \"INSERT.*INTO kanban_notify_subs\"). claim_unseen_events_for_sub / advance_notify_cursor / rewind_notify_cursor all read/write the existing cursor and need no change. No widening needed.

Related Issue

Fixes #29905

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor
• 🎯 New skill

Changes Made

• hermes_cli/kanban_db.py — add_notify_sub now snaps last_event_id to the current MAX(id) over task_events for the task after a newly-inserted row.
• tests/hermes_cli/test_kanban_notify.py — two new regression tests: backlogged terminal events must NOT replay for a newly-created sub; fresh post-subscription events MUST still deliver; idempotent re-subscription MUST NOT rewind the cursor.
• tests/hermes_cli/test_kanban_notify.py::test_notifier_skips_subscription_owned_by_other_profile — invariant updated from "cursor must equal 0" to "cursor must equal the snapped baseline" (same intent: wrong-profile watcher must not advance the cursor).
• tests/hermes_cli/test_kanban_core_functionality.py::test_notify_claim_is_single_owner_and_rewindable — old_cursor baseline is now the snapped value rather than the hard-coded 0; the rest of the single-owner / rewind invariants are unchanged.

How to Test

uv run --with pytest --with pytest-xdist --with pytest-asyncio --with pytest-timeout \
  python3 -m pytest tests/hermes_cli/test_kanban_notify.py \
                    tests/gateway/test_kanban_notifier.py \
                    tests/hermes_cli/test_kanban_db.py \
                    tests/hermes_cli/test_kanban_core_functionality.py -v

Manual repro per the issue: create a task, append a couple of terminal events directly to task_events, then call add_notify_sub — on main the next watcher tick fires once per backlogged event; with this PR no notifications fire until a new event arrives.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(kanban):)
• I searched for existing PRs to make sure this isn't a duplicate (no open/merged/closed PR by anyone addresses the storm-on-boot cursor-snap; feat: board-level kanban notification subscriptions (ADR-0011) #28748 / feat(kanban): auto-subscribe, rerun command, and test coverage #29891 / feat(kanban): auto-subscribe on task creation + rerun command + parent workspace #29097 are orthogonal feature work on board-level subs and auto-subscribe)
• My PR contains only changes related to this fix
• I've run focused tests for the touched code and all pass
• I've added tests for my changes
• I've tested on my platform: macOS 15.x

Documentation & Housekeeping

• I've updated relevant documentation — N/A (internal DB helper; no public-facing config or docs key changed)
• I've updated cli-config.yaml.example if I added/changed config keys — N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — N/A (pure SQLite/Python logic)
• I've updated tool descriptions/schemas if I changed tool behavior — N/A

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR prevents notification subscriptions from replaying backlogged terminal task events by snapping new subscription cursors to the current task_events max id, and updates/extends tests to assert the new cursor semantics (including idempotent re-subscribe behavior).

Changes:

• Update add_notify_sub to initialize last_event_id for newly-inserted subscriptions to MAX(task_events.id) for the task.
• Adjust existing tests that previously assumed last_event_id == 0 to instead assert against the snapped cursor baseline.
• Add regression tests covering “late subscribe” catch-up behavior and idempotent subscriptions not rewinding the cursor.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
hermes_cli/kanban_db.py	Snap new notify subscription cursor to MAX(task_events.id) to avoid replay storms.
tests/hermes_cli/test_kanban_notify.py	Add/adjust tests validating cursor snapping and idempotent behavior.
tests/hermes_cli/test_kanban_core_functionality.py	Update expectation to use snapped cursor as the baseline instead of 0.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[briandevans]

@copilot All three findings addressed in commit e7727de8c:

• tests/hermes_cli/test_kanban_notify.py:663,689,715 — _append_event calls now wrapped in with kb.write_txn(conn): per the function's documented "called from within an already-open txn" contract.
• tests/hermes_cli/test_kanban_notify.py:672 — sub["last_event_id"] cast to int before == max_id_before.
• tests/hermes_cli/test_kanban_notify.py:724,730 — sub_before/sub_after["last_event_id"] cast to int before numeric comparison, matching the convention used elsewhere in this file.

The two failing CI tests (test_browser_secret_exfil::test_allows_normal_url and test_browser_supervisor::test_supervisor_start_and_snapshot) are unrelated network/Chrome environment failures — they live in tests/tools/ and have no dependency on the kanban DB path touched by this PR.

[briandevans]

Closing to keep my queue focused — this P3 kanban notify-cursor snap has been idle ~22 days with no maintainer engagement and the kanban SQLite area is actively owned by other contributors. Happy to reopen if maintainers want it. Thanks!