wyoming_openai: Skill for Claude Code

Current State

• Branch: !git branch --show-current
• Dirty files: !git diff --name-only

Run all project linters and resolve every issue. Loop until the codebase passes cleanly.

Python environment

Use the repository's selected or activated virtual environment for every linter invocation. Do not assume a particular environment name or location. Resolve the interpreter from the active environment (for example, with python -c "import sys; print(sys.executable)") and refer to it below as <venv-python>.

Run Ruff and Pyright as modules through <venv-python> rather than using system-level commands. Pass the same interpreter to Pyright with --pythonpath so its import resolution uses the environment containing the project's dependencies. If no project environment is selected, stop and report that linting requires the development environment to be activated or selected.

Linter pipeline

Run the linters in this order each iteration:

1. Ruff auto-fix — <venv-python> -m ruff check . --fix (safe, auto-fixable issues)
2. Ruff check — <venv-python> -m ruff check . (remaining issues that need manual fixes)
3. Pyright — <venv-python> -m pyright --pythonpath <venv-python> (type-checking)

Resolution loop

while linters report errors:
    1. Run <venv-python> -m ruff check . --fix  (let ruff auto-fix what it can)
    2. Run <venv-python> -m ruff check .
       - If errors remain, read the offending files and fix them
    3. Run <venv-python> -m pyright --pythonpath <venv-python>
       - If errors remain, read the offending files and fix them
    4. If no errors in steps 2 and 3, break

Rules

• Max 5 iterations. If issues persist after 5 loops, stop and report the remaining errors to the user.
• Ruff first, pyright second. Ruff fixes (especially import sorting) can eliminate pyright false positives, so always run ruff before pyright.
• Use ruff check . --fix before manual fixes. Never manually fix what ruff can auto-fix.
• Minimal changes. Only touch lines the linters flag. Do not refactor, add docstrings, or "improve" surrounding code.
• Preserve behavior. Fixes must not change runtime behavior. If a fix requires a judgment call (e.g., an Any type that needs a real annotation), prefer the narrowest correct type.
• Report clearly. After each iteration, briefly state which linter you ran and how many errors remain. When done, summarize what was fixed.
• If $ARGUMENTS contains a file path or glob, pass it to both linters to scope the run (e.g., <venv-python> -m ruff check $ARGUMENTS and <venv-python> -m pyright --pythonpath <venv-python> $ARGUMENTS).