fix(memory): pass user_id through prefetch + scope per-user (Codex PR#33 P1)

[dizhaky]

Summary

• MemoryProvider.prefetch() base class: add user_id kwarg (all providers now have a consistent signature)
• All 9 plugins updated: mem0, honcho, hindsight, memgw, holographic, retaindb, supermemory, openviking, byterover
• memgw: scope prefetch results by user_id — prevents cross-user memory leakage in gateway sessions
• mem0: use per-turn user_id in sync_turn
• memory_manager.prefetch_all(): threads user_id through to all providers
• conversation_loop: passes agent._user_id to prefetch_all
• gateway/run.py: refreshes agent identity on cached-agent reuse (prevents stale identity after session swap)
• Tests updated to match new prefetch(user_id=) signature

Supersedes #36 and #39 (original commits were unsigned cursor-agent work; squashed into one signed commit).

Fixes Codex PR#33 P1 finding — cross-user identity + prefetch scoping.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes memory recall/write scoping and gateway cached-agent identity—important for multi-user sessions; most providers only extend signatures, with memgw/mem0 carrying the behavioral fixes.

Overview
Fixes cross-user memory identity in gateway sessions by plumbing user_id through the memory prefetch path and tightening two provider-specific behaviors.

MemoryProvider.prefetch() and MemoryManager.prefetch_all() now accept user_id; the conversation loop passes agent._user_id into prefetch_all so per-turn recall can be scoped to the current caller. All memory plugins adopt the same prefetch(..., user_id=) signature for consistency.

memgw tags background prefetch results with the requesting user_id and drops cached prefetch when prefetch is called for a different user, avoiding stale cross-user injection in shared sessions. mem0 sync_turn writes with user_id from the turn (falling back to config) instead of only static write filters.

When the gateway reuses a cached agent, it now refreshes _user_id, chat/thread metadata, and related identity fields from the current source, so shared-thread sessions do not keep the previous turn’s caller after a swap.

^{Reviewed by Cursor Bugbot for commit 9207fd1. Configure here.}

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[github-actions]

🔎 Lint report: dan/codex-p1-user-id-scoping 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: 8649 on HEAD, 8649 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 4570 pre-existing issues carried over.

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

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{Bugbot Autofix is ON, but it could not run because the branch was deleted or merged before autofix could start.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 9207fd1. Configure here.}

[cursor]

Mem0 prefetch ignores per-turn user

High Severity

The prefetch mechanism has inconsistent user_id handling. queue_prefetch may search using an outdated user ID, and the prefetch method's cross-user guard can fail if the incoming user_id is empty. This can result in incorrect or leaked memories when agents are reused across different users.

Additional Locations (1)

• plugins/memory/memgw/__init__.py#L290-L298

 

^{Reviewed by Cursor Bugbot for commit 9207fd1. Configure here.}

[dizhaky]

Addressed in two merged follow-up PRs:

• PR fix(mem0): scope prefetch search and result cache by user_id #44 (fix(mem0): scope prefetch search and result cache by user_id) — queue_prefetch now passes the kwarg user_id to Mem0 search; the cross-user guard also handles empty user_id via or not cached[1].
• PR fix(memory): pass per-turn user_id through model tool calls + compat fallback #48 (fix(memory): pass per-turn user_id through model tool calls + compat fallback) — all Mem0 tool call branches use per-call filters derived from the user_id kwarg, eliminating the stale self._user_id read.

Reviewed and confirmed by Claude Code.

---

Generated by Claude Code