agent: harden runtime reliability

[G9000]

Summary

• Isolate companion conversation history by thread and reload runtime history through explicit thread IDs.
• Serialize approval resume under the same per-thread lock as normal turns and preserve cancelled runs as terminal, including task cancellation cleanup.
• Scope delegated action results to their originating connection, preserve OpenAI-compatible temperature through tool binding, and align prompt wording with strict tool schemas.
• Add the completed runtime reliability plan and update the v2 memory recall scratchboard todo.

Test Plan

• bun run test:server (1397 passed, 1 skipped)
• bun run lint
• bun run build
• git diff --check

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e4091d16f5

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Pull request overview

Hardens the agent runtime’s correctness under concurrency by tightening per-thread state isolation, ensuring approval re-entry and cancellation are serialized/terminal-safe, and scoping delegated tool results to the originating WebSocket connection. It also aligns prompts/tool binding behavior with strict tool schemas and preserves user-scoped DB routing in background consolidation/episode generation paths.

Changes:

• Isolate companion conversation history per thread and refresh caches explicitly by thread_id.
• Serialize approval resume under the per-thread lock; preserve cancellation as terminal (including task-cancel cleanup).
• Scope delegated tool results by connection, preserve temperature through OpenAI-compatible tool binding, and remove schema-invalid “thinking argument” prompt wording.

Reviewed changes

Copilot reviewed 20 out of 20 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
scratchboard/v2-memory-recall-reliability/todo.md	Records Phase 4 runtime reliability work + validation gates.
docs/superpowers/plans/2026-05-18-agent-runtime-reliability.md	Adds the implementation plan covering AR-001..AR-006.
apps/server/tests/test_p5_transcript_archive.py	Adds coverage ensuring consolidation uses user-scoped soul DB factories.
apps/server/tests/test_inner_thought.py	Updates assertions to match prompt wording (no required thinking arg).
apps/server/tests/test_client_action_tools.py	Adds regression ensuring duplicate tool_call_ids are scoped by connection.
apps/server/tests/test_approval_reentry.py	Adds regression proving approval resume takes the per-thread lock.
apps/server/tests/test_agent_system_prompt.py	Asserts prompt no longer references a “thinking argument”.
apps/server/tests/test_agent_service.py	Adds regression tests for thread-scoped history + cancellation terminal behavior.
apps/server/tests/test_agent_openai_compatible_client.py	Adds regression ensuring bind_tools() preserves temperature.
apps/server/tests/test_agent_episodes.py	Adds regression ensuring episode generation uses user-scoped DB factory.
apps/server/src/anima_server/services/agent/templates/system_rules.md.j2	Removes schema-invalid “thinking argument” requirement wording.
apps/server/src/anima_server/services/agent/templates/system_prompt.md.j2	Rewords cognitive loop to avoid requiring schema-absent args.
apps/server/src/anima_server/services/agent/service.py	Adds thread lock around approval resume; thread-scoped cache refresh; cancellation cleanup on task cancel.
apps/server/src/anima_server/services/agent/persistence.py	Makes finalize_run() refuse to overwrite cancelled/failed runs.
apps/server/src/anima_server/services/agent/openai_compatible_client.py	Preserves temperature when binding tools.
apps/server/src/anima_server/services/agent/episodes.py	Switches to get_user_session_factory(user_id) for soul DB access.
apps/server/src/anima_server/services/agent/eager_consolidation.py	Uses user-scoped soul DB factories per thread in consolidation/sweeps.
apps/server/src/anima_server/services/agent/companion.py	Stores conversation windows per thread and returns defensive history copies.
apps/server/src/anima_server/services/agent/client_actions.py	Keys pending delegated tool calls by connection identity + call id; validates tool_name on resolve.
apps/server/src/anima_server/api/routes/ws.py	Resolves delegated tool results via the authenticated connection context.

---

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

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

apps/server/src/anima_server/services/agent/client_actions.py:135

• ClientActionRegistry now scopes pending delegated tool calls by (id(conn), tool_call_id), but tool_call_id can be deterministic (e.g., AgentRuntime generates synthetic-{tool}-{step_index}-{i}), so two concurrent runs on the same connection can still collide and overwrite each other’s pending entry. This can route a tool_result to the wrong awaiting turn. Consider namespacing the wire tool_call_id with run_id/thread_id (or generating a server-side UUID and mapping back) so it’s unique per connection even when models reuse IDs.

        if conn is None:
            raise DelegationTimeout(
                f"No connected client has registered action tool {tool_name!r}"
            )

        loop = asyncio.get_running_loop()
        future: asyncio.Future[DelegatedToolResult] = loop.create_future()
        pending_key = (id(conn), tool_call_id)
        self._pending[pending_key] = _PendingActionCall(
            connection=conn,
            future=future,
            tool_name=tool_name,
        )

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 21 out of 21 changed files in this pull request and generated 2 comments.

[Copilot]

Just as a heads up, I was blocked by some firewall rules while working on your feedback. Expand below for details.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• invalid
  • Triggering command: .venv/bin/pytest .venv/bin/pytest apps/server/tests -q /home/REDACTED/work/animaOS/animaOS/target/release/build/zstd-sys-496f82dd5cfa2083/out/fb80479a5fb81f6a-hist.o /home/REDACTED/work/animaOS/animaOS/target/release/build/zstd-sys-496f82dd5cfa2083/out/fb80479a5fb81f6a-huf_compress.o /home/REDACTED/work/animaOS/animaOS/target/release/build/zstd-sys-496f82dd5cfa2083/out/fb80479a5fb81f6a-zstd_compress.o /home/REDACTED/work/animaOS/animaOS/target/release/build/zstd-sys-496f82dd5cfa2083/out/fb80479a5fb81f6a-zstd_compress_literals.o /home/REDACTED/work/animaOS/animaOS/target/release/build/zstd-sys-496f82dd5cfa2083/out/fb80479a5fb81f6a-zstd_compress_sequences.o /home/REDACTED/work/animaOS/animaOS/target/release/build/zstd-sys-496f82dd5cfa2083/out/fb80479a5fb81f6a-zstd_compress_superblock.o /home/REDACTED/work/animaOS/animaOS/target/release/build/zstd-sys-496f82dd5cfa2083/out�� /home/REDACTED/work/animaOS/animaOS/target/release/build/zstd-sys-496f82dd5cfa2083/out/fb80479a5fb81f6a-zstd_fast.o /home/REDACTED/work/animaOS/animaOS/target/release/build/zstd-sys-496f82dd5cfa2083/out/fb80479a5fb81f6a-zstd_lazy.o know�� known-linux-gnu/-m64 known-linux-gnu/-I stable-x86_64-unzstd/lib/ S/target/releasear S/target/releasecqD ld.lld S/target/release/home/REDACTED/work/animaOS/animaOS/target/release/build/zstd-sys-496f82dd5cfa2083/out/44ff4c55aa9e5133-debug.o (dns block)
  • Triggering command: /home/REDACTED/work/animaOS/animaOS/.venv/bin/pytest pytest stable-x86_64-un-c stable-x86_64-un-- stable-x86_64-un"/home/REDACTED/work/animaOS/animaOS/.venv/lib/python3.12/site-packages/pgserver/pginstall/bin/postgres" --single -F -O -j -c search_path=pg_catalog -c exit_on_er--norc 1a.r�� 496f82dd5cfa2083embed-bitcode=no S/target/release--allow=non_upper_case_globals /deps/rustce1Qka--cfg /deps/thiserror_iptables /deps/thiserror_-w /deps/thiserror_-t /deps/thiserror_security /dep�� /deps/thiserror_OUTPUT kages/pgserver/p-d /deps/thiserror_168.63.129.16 /deps/thiserror_bash /deps/thiserror_--norc /deps/thiserror_--noprofile /deps/thiserror_owner (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

[Copilot]

Pull request overview

Copilot reviewed 22 out of 22 changed files in this pull request and generated no new comments.