| name | axrl-fix-linting |
| description | Fix lint, formatting, pyright, and mypy failures in this repository by running scripts/run-precommit-check.sh, making the smallest non-semantic changes, and stopping if a fix would change logits or training semantics. |
| argument-hint | Describe the failing check, files, or error messages to fix |
| user-invocable | true |
axrl-fix-linting
Use this skill to fix lint, formatting, pyright, or mypy failures in this repository with the standard validation command.
Defaults
- Use
bash scripts/run-precommit-check.sh as the source of truth.
- Run the lint/type-check workflow on the local CPU node in the shared workspace, not on GPU nodes. GPU nodes may lack outbound network access for pre-commit hook initialization, and the local CPU node shares the same files and Docker image.
- Keep fixes minimal and non-semantic by default.
- Do not change logits, training semantics, reward logic, sampling behavior, or public behavior unless the user explicitly approves it.
- Use the existing Python environment if present; do not create a new virtual environment unless the user asks.
The repo check command is:
bash scripts/run-precommit-check.sh
It runs pre-commit, pyright, and mypy.
Workflow
- Run the repo check script locally on the CPU node and fix the first failing stage.
- Prefer the smallest local fix:
- formatter or lint failure: style-only fix
- pyright or mypy failure: fix the owning type contract or shared source
- After the first edit, run the narrowest relevant check, then rerun
bash scripts/run-precommit-check.sh.
- If a fix would change behavior or logits, stop and ask before proceeding.
- If several failures share one root cause, fix the shared source once instead of scattering one-off patches.
Done When
bash scripts/run-precommit-check.sh passes, or- the remaining blocker is identified clearly, including whether it is style-only, type-only, or semantic.
