The test was picking up the real OPENAI_API_KEY from the shell
environment, causing it to fail on any machine with that key set.
Added @patch.dict(os.environ, {}, clear=True) so the test runs in
a clean env and exercises the file_env path as intended.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: fix YAML error in argument-hint
* add codex auth support to responses API
* Use gpt-5.1-codex-mini as default model for Codex auth
Add CODEX_FALLBACK_MODELS chain (gpt-5.1-codex-mini → gpt-5.2) for
Codex endpoint which doesn't support standard OpenAI models like
gpt-4o-mini. Adds model fallback retry on 400 errors in the Codex
search path. Also adds test_codex_auth.py with 22 unit tests covering
JWT decoding, auth resolution, SSE parsing, and payload building.
* Pass .env credentials to Bird Node subprocesses for X auth
On platforms without browser cookie access (e.g. WSL2), Bird's
vendored Node.js module cannot read AUTH_TOKEN/CT0 from Firefox
or Chrome cookie stores. The .env config file already supports
these values, but they were only loaded into the Python config
dict — never exported to the environment of Node subprocesses.
- Add AUTH_TOKEN/CT0 to env.py config key loading
- Add set_credentials()/\_subprocess_env() to bird_x.py to inject
credentials into the env dict passed to subprocess.run/Popen
- Call set_credentials() in main() before Bird auth detection
---------
Co-authored-by: Justin Williams <jblwilliams@gmail.com>