bug: A2A endpoint returns NoneType path error — all skills failing

[mabry1985]

Summary

Quinn's A2A endpoint is failing on all skill invocations with:

Error: expected str, bytes or os.PathLike object, not NoneType

This affects all skills (bug_triage, pr_review, board_audit, etc.).

Impact

• bug_triage: 24% success rate (25 failures / 33 attempts)
• auto-triage-sweep: 0% success rate across 23 outcomes
• 25 items stuck in error state on the workstacean flow board
• Every fix_ci remediation dispatched by pr-remediator to Quinn silently fails
• Quinn is effectively offline for all A2A skill calls

Reproduction

Invoke Quinn via A2A JSON-RPC with any skill — response is always the NoneType path error.

Likely Cause

A file path variable (config path, workspace path, or repo clone path) is None when Quinn tries to use it. Candidates:

• Missing environment variable for workspace/temp directory
• Repo path not resolved before use in skill handler entry point
• Config file path refactored but not updated

Quinn's own auto-triage confirmed: "this is a runtime/deployment issue, not a build failure. CI is green across all recent runs on dev."

Cross-reference from workstacean side:

• bug: Quinn A2A endpoint returns NoneType path error — bug_triage at 24% success rate protoWorkstacean#332 (original filing + triage comments)
• bug: Quinn board_audit and pr_review skills fail with PathLike TypeError protoWorkstacean#339 (duplicate — also noted protoMaker agent returns 404 on chat, possible shared config root cause)

Suggested Fix

1. Check Quinn's skill handler entry point for Path() or open() calls receiving args from config/env
2. Add null guards before path operations
3. Verify all required env vars in Quinn's deployment config (especially workspace/temp dirs)

Priority

Critical — this blocks all automated triage and QA workflows across the fleet.

[mabry1985]

Diagnostic gap

Can't fully diagnose this from the current report — the root-cause candidates (gh-tool bypass, Path(None), env var) all look guarded in current `dev`:

• `pr_inspector`, `github_actions`, `github_issues`, `release_notes` each require explicit `repo` and return a readable guard error on missing/blank (`tests/test_gh_tools_repo_required.py` locks this in — all pass).
• The two LangGraph entry points (`_chat_langgraph`, `_chat_langgraph_stream`) both wrap in `try/except` and call `log.exception(...)` so a real `TypeError` would land in `docker logs quinn` with a full traceback (`tests/test_exception_logging.py` locks this in).

To unblock: please paste the docker logs quinn lines around one failed skill invocation. Specifically looking for:

```
ERROR quinn.server [quinn-a2a-stream] unhandled exception for session=...
Traceback (most recent call last):
...
TypeError: expected str, bytes or os.PathLike object, not NoneType
```

If no such line exists, the error is being raised and swallowed somewhere we don't have a logger attached yet — that itself is the first fix.

If the container isn't logging at INFO: check `LOG_LEVEL` — root config reads it from env, defaults to INFO (`server.py:33-36`). If the deploy overrides to WARNING, the exception logger still fires (it's ERROR) but other context disappears.

Cross-reference: if workstacean's side has captured the raw A2A response body on a failed dispatch, the `status.message` text field carries `str(exc)` from Quinn — paste that with the trace_id and we can pull the Langfuse trace.