fix: preserve codename on re-registration after stale cleanup

[evansenter]

Summary

• When a session's heartbeat exceeded the 24h timeout, cleanup_stale_sessions() would soft-delete it. On re-registration, find_session_by_client() filtered out deleted sessions, causing a new codename to be generated
• Added include_deleted parameter to find_session_by_client() to recover soft-deleted sessions during registration, reactivating by clearing deleted_at
• Added 5 tests covering codename preservation, session reactivation, and active-session preference in lookups

Test plan

• Existing 299 tests pass
• New regression test: codename preserved after soft-delete + re-register
• New test: deleted_at cleared on reactivation
• New storage tests: include_deleted parameter behavior (default excludes, flag includes, prefers active)

Closes #104

🤖 Generated with Claude Code

[claude]

Prompt: evansenter/dotfiles/.../claude-review.md

Code Review

Summary

Fixes codename loss after stale session cleanup by adding include_deleted to find_session_by_client() so that re-registration with the same (machine, client_id) recovers the soft-deleted session's codename and reactivates it. Includes 5 new tests covering the regression scenario and storage-level behavior.

Issues Found

Critical

None

Important

• server.py:217-220 - Session reactivation does not update cwd or repo. When a session is reactivated, only name, last_heartbeat, and deleted_at are updated on the existing session object before calling add_session(). The cwd and repo values passed to register_session() are returned in the API response (lines 231-232) but the stale values from the original session are persisted to the database. This means list_sessions and other storage queries will show the old cwd/repo, while the caller received the new ones. The fix is to also set existing.cwd = cwd and existing.repo = repo before calling add_session(). (Note: this is a pre-existing issue for the active-session resume path, but this PR widens the window since sessions can now be recovered after potentially much longer periods where cwd is more likely to have changed.)

Suggestions

• storage.py:430 - The SQL ORDER BY deleted_at IS NOT NULL, last_heartbeat DESC correctly prefers active sessions over deleted ones, but the intent relies on SQLite's boolean-as-integer ordering (0 for NULL/active, 1 for NOT NULL/deleted). A comment explaining this ordering would help future readers.
• test_storage.py:228-249 - The test_find_session_by_client_include_deleted_prefers_active test creates two sessions with the same (machine, client_id) but different id values. The comment acknowledges "This shouldn't normally happen." Since the (machine, client_id) index is not UNIQUE, this scenario is technically possible but might be worth a brief note on how it could happen (e.g., a race condition or manual DB edit), to make the test's purpose clearer to future readers.

Verdict

REQUEST_CHANGES - One important issue: session reactivation should also update cwd and repo on the persisted session object to avoid a mismatch between the API response and stored state.

---

Automated review by Claude Code

[evansenter]

Feedback Addressed

Implemented

• [Important] server.py:217-220 - Session reactivation now updates cwd and repo on the existing session object before add_session(), keeping stored state consistent with the API response
• [Suggestion] storage.py:430 - Added inline comment explaining SQLite boolean-as-integer ordering (0 for active, 1 for deleted) in the ORDER BY clause
• [Suggestion] test_storage.py:232-233 - Updated test comment to explain how duplicate (machine, client_id) entries could occur (race conditions, manual DB edits)

[claude]

Prompt: evansenter/dotfiles/.../claude-review.md

Code Review

Summary

Fixes codename loss after stale session cleanup by adding include_deleted to find_session_by_client() so that re-registration recovers soft-deleted sessions. Second round addresses all prior feedback: cwd/repo are now updated on reactivation, SQL ordering is commented, and the edge-case test comment is clarified.

Issues Found

Critical

None

Important

None

Suggestions

None

Previously Addressed (Filtered)

• server.py:217-220 - Session reactivation now updates cwd and repo (Implemented)
• storage.py:430 - Added inline comment explaining SQLite boolean-as-integer ordering (Implemented)
• test_storage.py:232-233 - Updated test comment to explain how duplicate entries could occur (Implemented)

3 items from prior feedback rounds were not re-raised.

Verdict

APPROVE - All prior feedback has been addressed. The fix correctly recovers codenames from soft-deleted sessions, updates all relevant session fields on reactivation, and includes thorough test coverage for the regression and storage-level behavior.

---

Automated review by Claude Code