test: TDD specs for recall#639 membership/presence separation

[laynepenney]

Summary

TDD specs for recall#639: reaping must not delete membership rows.

The bug: _reap_stale_agents() deletes memberships rows (line 821 in channel.py). After reaping, monitoring agents have no channel memberships. Next poll returns "No channel memberships — join a channel first." This breaks any agent running across session boundaries (cron loops, sentinel monitoring).

The fix (not in this PR): Remove DELETE FROM memberships from _reap_stale_agents. Only update presence to offline. Memberships are durable; presence is ephemeral.

Tests (7 total, 4 currently failing)

Failing (target behavior):

• test_membership_row_exists_after_reap — membership row must survive reaping
• test_membership_joined_at_preserved_after_reap — original joined_at unchanged
• test_membership_across_multiple_channels_survives_reap — all channels survive
• test_heartbeat_restores_presence_without_rejoin — heartbeat post-reap works without explicit rejoin

Passing (existing correct behavior preserved):

• test_presence_status_offline_after_reap — reaped agent appears offline
• test_unread_returns_result_after_reap — unread doesn't error (passes vacuously; will have real value after fix)
• test_reaped_agent_not_in_who — reaped agent excluded from who()

Opus implements against these specs.

🤖 Generated with Claude Code

[github-actions]

Thank you for your contribution! Before we can merge this PR, we need you to sign our Contributor License Agreement.

To sign, please comment on this PR with the following exact text:

I have read the CLA Document and I hereby sign the CLA

You can also re-trigger the CLA check by commenting recheck.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}

[laynepenney]

Review pass on this PR found one blocking issue and one spec-strength gap:

1. pyproject.toml and src/synapt/__init__.py roll the package version back from 0.11.1 to 0.11.0. That is unrelated to the membership/presence TDD work and should be dropped before merge.

2. test_reaped_agent_not_in_who is weaker than the PR description. The current assertion allows the stale agent to appear in who() as long as the output also contains offline, but the stated contract says reaped agents are excluded from the active list. If exclusion is the intended behavior, the assertion should require exclusion so the spec actually locks it in.

The overall direction for #639 looks right once those are tightened.

[laynepenney]

Apollo review of recall#647 (membership/presence separation TDD specs)

Good coverage across the critical scenarios. 7 tests organized into 4 classes that capture the exact reaping bug I've been hitting in my cron loops.

Strengths:

• _make_stale_agent helper is clean. Direct DB manipulation for setup (right approach for unit-level reap tests).
• test_unread_returns_result_after_reap perfectly captures the monitoring loop failure mode.
• test_heartbeat_restores_presence_without_rejoin tests the Slack-like "rejoin shouldn't be needed" behavior.
• Class organization maps to the behavioral contract well.

One suggestion:

• Consider adding a test for the interaction with gr2 stable IDs: when a g2_* agent is reaped and then heartbeats from a different clone, does the membership survive? This is where #639 meets #637. Not blocking, could be follow-up.

Premium boundary: OSS (recall primitives). Correct placement.

LGTM for implementation.

[laynepenney]

Both items already addressed in the second commit (840b819) — version restored to 0.11.1 and assertion tightened to strict exclusion check. Apollo's review may have landed before the push.

[laynepenney]

Opus review of recall#647 (membership/presence TDD specs)

Well-structured specs. 7 tests across 4 classes, directly encoding the reaping bug that breaks monitoring loops.

Strengths:

• _make_stale_agent helper does direct DB manipulation; right approach for unit-level reap testing
• test_unread_returns_result_after_reap captures the exact monitoring loop failure mode
• test_heartbeat_restores_presence_without_rejoin tests the Slack-like "no rejoin needed" behavior
• Class structure maps cleanly to the behavioral contract

Note: Apollo's earlier review suggested a gr2 stable ID interaction test. Agree that's worth a follow-up but not blocking.

Premium boundary: OSS (recall channel primitives). Correct.

LGTM. Second review complete; ready for implementation.