🔎 Lint report: hermes/hermes-82ce03d4 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 10895 on HEAD, 10893 on base (🆕 +2)

🆕 New issues (2):

tests/run_agent/test_credits_notices_toggle.py:76: [unresolved-attribute] unresolved-attribute: Unresolved attribute `_credits_session_start_micros` on type `AIAgent`
run_agent.py:2920: [unresolved-attribute] unresolved-attribute: Object of type `Self@get_credits_spent_micros` has no attribute `_credits_session_start_micros`

✅ Fixed issues (1):

tests/run_agent/test_credits_notices_toggle.py:76: [invalid-assignment] invalid-assignment: Object of type `None` is not assignable to attribute `_credits_session_start_micros` of type `int`

Unchanged: 5724 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.

Verification: LGTM — solid identity-based flush fix

Reviewed the full diff. The positional-slice approach (messages[flush_from:]) was fragile when repair_message_sequence merged/shrunk the history — len(conversation_history) > len(messages) made the slice empty, silently dropping assistant responses (#46053).

The replacement using id(msg) identity tracking is correct:

• History dicts are skipped by identity (same object in both messages and conversation_history)
• New dicts appended during the turn are written once, even if repair compacts the list around them
• Repeated flush calls preserve bug: SQLite session transcript accumulates duplicate messages (3-4x token inflation) #860 dedup behavior via the flushed_ids set
• Session-id scoping resets the identity window on session rotation

Test coverage is thorough — 4 new tests in test_identity_flush.py covering the exact regression scenario (stale cursor, repeated flush, cursor reset on new turn). The existing test_flush_with_stale_history_loses_messages is updated from "verifies the bug exists" to "verifies the fix works" — clean transition.

One note: _last_flushed_db_idx is still set to len(messages) at the end (line ~1633), but it's no longer used for slicing — only for the session-id reset gate. This is harmless but could be cleaned up in a follow-up if desired.

Live testing added after CI because this touches the core persistence path:

The repair-flow test is the sensitive case from #46053: persisted conversation_history is longer than the repaired working messages list. The assistant row still landed in state.db.