feat(cli): add result metadata fd transport - #18
Conversation
Refs cermm/wc-infrastructure#358 Refs cermm/wc-infrastructure#401
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 18b50851f0
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".
| if descriptor_target is None: | ||
| raise ResultMetadataError( | ||
| "result metadata descriptor identity is unavailable" | ||
| ) | ||
| if not descriptor_target.startswith("pipe:"): |
There was a problem hiding this comment.
On macOS and other POSIX hosts without Linux procfs, /proc/self/fd is unavailable and /dev/fd/<n> is not guaranteed to be a symlink whose target begins with pipe:. Consequently, a valid anonymous pipe that already passes fstat, access-mode, blocking, and PC_PIPE_BUF validation is rejected as having unavailable or invalid identity, despite the option being documented for POSIX generally. The fallback test only mocks /proc failure while retaining Linux /dev/fd behavior, so this needs validation on the actual supported host path.
AGENTS.md reference: AGENTS.md:L84-L87
Useful? React with 👍 / 👎.
| if getattr(self, "result_meta_fd", None) is not None: | ||
| self._publish_result_metadata(result) |
There was a problem hiding this comment.
This publication point is skipped whenever chat() returns before reaching the agent result, including _ensure_runtime_credentials() or _init_agent() returning false; the quiet path has the same omission when credential or agent initialization fails. The ownership guards then close the accepted descriptor without writing JSON, and the nonquiet path can even return success with an empty pipe, breaking callers that rely on the advertised one-frame contract. Unexpected exceptions in the nonquiet caller are likewise not converted because that caller catches only KeyboardInterrupt; centralize a bounded failure-frame fallback across both query paths.
AGENTS.md reference: AGENTS.md:L54-L57
Useful? React with 👍 / 👎.
| if type(max_iterations) is not int or max_iterations < 0: | ||
| max_iterations = 90 | ||
| upper_bound = min(max_iterations + 1, MAX_API_CALLS) | ||
| value = result.get("api_calls") | ||
| if "api_calls" in result and type(value) is int and 0 <= value <= upper_bound: |
There was a problem hiding this comment.
With the normal max_turns=90, successful turns may legitimately report 33–91 API calls, but this fixed cap rejects every count above 32, replaces it with zero, and changes failure_class from none to unknown_failure even when completed is true. This corrupts the metadata precisely for longer tool-driven runs; the accepted bound should remain related to max_iterations + 1 rather than freezing an unrelated enumeration limit.
AGENTS.md reference: AGENTS.md:L80-L83
Useful? React with 👍 / 👎.
|
Closing superseded PR; #19 carries the same reviewed result-metadata FD tree with accepted contributor attribution. |
1 participant
Summary
Implements the NousResearch#358/NousResearch#401 producer-bound
--result-meta-fdtransport on currentcermm/hermes-agentmain.hermes-agent-result-meta-v1frame generation.interruptedandunknown_failureframes for quietKeyboardInterrupt/ unexpected exceptions without raw exception/path/secret leakage.--result-meta-fdoptions before descriptor claim/startup and keeps the standalone parser aligned.Exact gate evidence
Base/fork main before candidate:
e0c3e1b098e719d2f7327b202755ace76414212f.Candidate commit:
18b50851f037c4686bd6be77aff092098973c899.Candidate tree:
bb1e4c3ada50641c5ecbf0a7260018f4bb76db5f.Reviewed Kanban gates:
t_f74bbba6repair handoff,t_2a4921d3independent GPT-5.6 PASS (PASS_REVIEW_RESULT_META_FD_DUPLICATE_REPAIR).Local verification rerun from
/home/michal/worktrees/nousresearch-hermes-agent/issue358-result-meta-current-main-20260816:scripts/run_tests.sh tests/hermes_cli/test_result_metadata.py tests/hermes_cli/test_result_metadata_cli.py -q→ 44 passed, 0 failed.test_result_metadata*, parser/flags/startup/safe-mode/relaunch/coalesce/container/yolo/subparser) → 209 passed, 0 failed./home/michal/.hermes/hermes-agent/venv/bin/python -m ruff check cli.py hermes_cli/_parser.py hermes_cli/main.py hermes_cli/result_metadata.py tests/hermes_cli/test_result_metadata.py tests/hermes_cli/test_result_metadata_cli.py→ All checks passed.python -m compileall -qon six Python scope files → exit 0.python -m hermes_cli.main chat --helpexposes--result-meta-fd.git diff --check→ exit 0.Boundaries
This is repository code/docs/tests only. It does not mutate live Hermes profiles, SOUL files, auth stores, runtime config, canary state, rollout state, or wc-infrastructure issue closure.
Refs cermm/wc-infrastructure#358
Refs cermm/wc-infrastructure#401