fix(agent): notify context_compressor on commit_memory_session

[wesleysimplicio]

Summary

AIAgent.commit_memory_session() is invoked when the session_id rotates
without tearing providers down — CLI /new, gateway session expiry, and
in-flight context compression. It already calls
_memory_manager.on_session_end(), but never called
context_compressor.on_session_end(), so plugin context engines like
hermes-lcm miss the final session-end flush:

• Messages that arrived after the last compress() call are never
  persisted to lcm.db.
• The LCM lifecycle store keeps the session marked active forever.

shutdown_memory_provider() already does both calls; commit_memory_session
was the asymmetric outlier.

Fix

Mirror the shutdown_memory_provider() pattern: the compressor flush is
now invoked under the same hasattr(...) and self.context_compressor
guard, and each backend is checked independently so a missing
_memory_manager no longer skips the compressor.

if self._memory_manager:
    try:
        self._memory_manager.on_session_end(messages or [])
    except Exception:
        pass
if hasattr(self, "context_compressor") and self.context_compressor:
    try:
        self.context_compressor.on_session_end(self.session_id or "", messages or [])
    except Exception:
        pass

Tests

New regression tests in tests/agent/test_commit_memory_session_compressor.py:

• test_notifies_both_memory_manager_and_context_compressor
• test_compressor_flushed_even_when_memory_manager_absent
• test_memory_manager_failure_does_not_block_compressor
• test_compressor_failure_is_swallowed
• test_no_op_when_neither_backend_present
• test_uses_empty_string_session_id_when_none

Stash-verified: 4/6 fail without the fix (the two no-op cases pass either
way), all 6 pass with it.

Test plan

• New regression tests pass with fix; fail without it
• Wider regression: pytest tests/agent/test_memory_provider.py tests/run_agent/ tests/cli/test_cli_new_session.py tests/agent/test_commit_memory_session_compressor.py → 1332 passed, 9 skipped
• Full regression: pytest → 21240 passed; the 115 failures match the pre-existing flaky baseline on origin/main (test_discord_, test_gateway_service, test_gateway_wsl, test_file_, test_terminal_tool_requirements, test_tts_media_routing, test_setup_openclaw_migration, test_tencent_tokenhub_provider, test_list_picker_providers, test_model_switch_custom_providers, test_tui_gateway_server, test_web_server::test_pub_broadcasts) — none touched by this PR

Closes #22394

[Copilot]

Pull request overview

This PR fixes an asymmetry in AIAgent.commit_memory_session() by ensuring it notifies the configured context engine (context_compressor) on session rotation, matching the behavior already implemented in shutdown_memory_provider(). This prevents plugin-based context engines (e.g., hermes-lcm) from missing the final session-end flush when sessions rotate without provider teardown (e.g., /new, gateway expiry, compression-driven rollover).

Changes:

• Update AIAgent.commit_memory_session() to independently flush _memory_manager and context_compressor (each guarded and exception-swallowed).
• Add regression tests covering fan-out behavior, failure isolation, and edge cases (missing backends, session_id=None).

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
run_agent.py	Ensures commit_memory_session() also calls context_compressor.on_session_end(...) under the same guards used elsewhere, and does not skip compressor flush when _memory_manager is absent.
tests/agent/test_commit_memory_session_compressor.py	Adds targeted regression tests verifying both backends are notified and failures in one backend do not block the other.

---

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

[alt-glitch]

Competing fix with #22431 — both address #22394. Same fix: notify context_compressor.on_session_end() in commit_memory_session().

[wesleysimplicio]

Closing this one to reduce queue noise. The underlying issue (#22394) was resolved via merged PR #22764, so this branch is now superseded.