Background review isolation and hindsight retain fixes

NousResearch/hermes-agent · cluster 47268 · agent

Data as of

Nothing open on GitHub. Fix merged in #27190. Check it covers your case.

The short answer

0open issues
0open PRs
5items grouped
Known?
Yes. 5 related items grouped here, 2 issues already closed.
Fix event
A fix was merged to main in PR #27190. Check it covers your case. No release data is tracked.
Tracker
The maintainers' tracker marks this unresolved. That is separate from the fix event.
Still open
Nothing open on GitHub.

These issues and PRs address bugs where background review processes incorrectly shared session context with the main agent loop, and fix destructive or unsafe behavior in hindsight memory retention. The work involves isolating background review forks and replacing unsafe retain operations with safer alternatives.

Open issues0

No open issues.

Open PRs0

No open pull requests.

Fixes on record

Merged to main. A stamp counts the issues a fix is linked to.

  • #27190 fix(run_agent): isolate background review fork from external memory plugins

    merged to main

    files: run_agent.py, tests/run_agent/test_background_review.py

    Show the diff
    • run_agent.py
      @@ -4386,6 +4386,21 @@ def _bg_review_auto_deny(command, description, **kwargs):
                           # owns the loop and the agent-loop tools dispatch.
                           if _parent_api_mode == "codex_app_server":
                               _parent_api_mode = "codex_responses"
      +                    # skip_memory=True keeps the review fork from
      +                    # touching external memory plugins (honcho, mem0,
      +                    # supermemory, etc.).  Without it, the fork's
      +                    # __init__ rebuilds its own _memory_manager from
      +                    # config, scoped to the parent's session_id, and
      +                    # run_conversation() then leaks the harness prompt
      +                    # into the user's real memory namespace via three
      +                    # ingestion sites: on_turn_start (cadence + turn
      +                    # message), prefetch_all (recall query), and
      +                    # sync_all (harness prompt + review output recorded
      +                    # as a (user, assistant) turn pair).  Built-in
      +                    # MEMORY.md / USER.md state is re-bound from the
      +                    # parent below so memory(action="add") writes from
      +                    # the review still land on disk; the review just
      +                    # has zero side effects on external providers.
                           review_agent = AIAgent(
                               model=self.model,
                               max_iterations=16,
      @@ -4397,6 +4412,7 @@ def _bg_review_auto_deny(command, description, **kwargs):
                               api_key=[redacted:secret-assignment]"api_key") or None,
                               credential_pool=[redacted:secret-assignment] "_credential_pool", None),
                               parent_session_id=self.session_id,
      +                        skip_memory=True,
                           )
                           review_agent._memory_write_origin = "background_review"
                           review_agent._memory_write_context = "background_review"
    • tests/run_agent/test_background_review.py
      @@ -193,3 +193,51 @@ def close(self):
           assert captured_bg_callback[0].startswith("💾 Self-improvement review:"), (
               captured_bg_callback[0]
           )
      +
      +
      +def test_background_review_fork_skips_external_memory_plugins(monkeypatch):
      +    """The background review fork must NOT touch external memory plugins.
      +
      +    Without skip_memory=True on the fork constructor, AIAgent.__init__
      +    rebuilds its own _memory_manager from config, scoped to the parent's
      +    session_id.  The review fork's run_conversation() then leaks the
      +    harness prompt into the user's real memory namespace via three
      +    ingestion sites: on_turn_start (cadence + turn message),
      +    prefetch_all (recall query), and sync_all (harness prompt + review
      +    output recorded as a (user, assistant) turn pair).  The fix is a
      +    single kwarg on the fork constructor — this test guards it.
      +    """
      +    captured_kwargs: dict = {}
      +
      +    class FakeReviewAgent:
      +        def __init__(self, **kwargs):
      +            captured_kwargs.update(kwargs)
      +            self._session_messages = []
      +
      +        def run_conversation(self, **kwargs):
      +            pass
      +
      +        def shutdown_memory_provider(self):
      +            pass
      +
      +        def close(self):
      +            pass
      +
      +    monkeypatch.setattr(run_agent_module, "AIAgent", FakeReviewAgent)
      +    monkeypatch.setattr(run_agent_module.threading, "Thread", ImmediateThread)
      +
      +    agent = _bare_agent()
      +
      +    AIAgent._spawn_background_review(
      +        agent,
      +        messages_snapshot=[{"role": "user", "content": "hello"}],
      +        review_memory=True,
      +    )
      +
      +    assert captured_kwargs.get("skip_memory") is True, (
      +        "Background review fork must be constructed with skip_memory=True "
      +        "so AIAgent.__init__ does not rebuild a _memory_manager wired to "
      +        "external plugins (honcho, mem0, supermemory, ...).  Without this "
      +        "the fork leaks harness prompts into the user's real memory "
      +        "namespace via on_turn_start / prefetch_all / sync_all."
      +    )

2 PRs closed without merging

From the close comments: done elsewhere 1 · other 1

  • #47364 fix(agent): stop bg-review from sharing parent session_id (#47268) · other → #47268
  • #55819 fix(hindsight): replace destructive aretain_batch with safe file-based retain · done elsewhere → 4ece521, #27190
Closed issues · 2
  • Closed#47268[Bug]: Background review shares session context with main loop, causing premature stop (11-77 char responses)
  • Closed#55816fix(hindsight): hindsight_retain destructive overwrite, misleading description, and Background Review abuse
How this cluster was grouped
  • Link evidence (partly structural links): some members are connected by structural evidence (shared changed files, or a closing/superseded reference); the rest are joined by an automated relevance judgement over their text.
  • Members outside that structural core may belong to a different underlying problem.
  • Pairwise link density 0.34: the share of item pairs that carry a link, counting a structural link in full and a relevance-only link at 0.36. 1.00 means every pair is linked.

Pick up a Lab task

Check whether a fix covers the open issues here, with your own agent. Bring back evidence for maintainers.

Open the Lab

At a glance

Subsystem
agent
Merged PRs
#27190
Closed issues
2
Attempts
2 PRs closed without merging
Fix link checked
17 May 2026

For your agent

The same facts, typed.