ruff-recursive-fix

SKILL.md

Ruff Recursive Fix

Overview

Use this skill to enforce code quality with Ruff in a controlled, iterative workflow.
It supports:

Optional scope limitation to a specific folder.

Default project settings from pyproject.toml.

Flexible Ruff invocation (uv, direct ruff, python -m ruff, or equivalent).

Optional per-run rule overrides (--select, --ignore, --extend-select, --extend-ignore).

Automatic safe then unsafe autofixes.

Diff review after each fix pass.

Recursive repetition until findings are resolved or require a decision.

Judicious use of inline # noqa only when suppression is justified.

Inputs

Collect these inputs before running:

target_path (optional): folder or file to check. Empty means whole repository.

ruff_runner (optional): explicit Ruff command prefix (for example uv run, ruff, python -m ruff, pipx run ruff).

rules_select (optional): comma-separated rule codes to enforce.

rules_ignore (optional): comma-separated rule codes to ignore.

extend_select (optional): extra rules to add without replacing configured defaults.

extend_ignore (optional): extra ignored rules without replacing configured defaults.

allow_unsafe_fixes (default: true): whether to run Ruff unsafe fixes.

ask_on_ambiguity (default: true): always ask the user when multiple valid choices exist.

Command Construction

Build Ruff commands from inputs.

0. Resolve Ruff Runner

Determine a reusable ruff_cmd prefix before building commands.

Resolution order:

If ruff_runner is provided, use it as-is.

Else if uv is available and Ruff is managed through uv, use uv run ruff.

Else if ruff is available on PATH, use ruff.

Else if Python is available and Ruff is installed in that environment, use python -m ruff.

Else use any project-specific equivalent that invokes installed Ruff (for example pipx run ruff), or stop and ask the user.

Use the same resolved ruff_cmd for all check and format commands in the workflow.

Base command:

<ruff_cmd> check

Formatter command:

<ruff_cmd> format

With optional target:

<ruff_cmd> format <target_path>

Add optional target:

<ruff_cmd> check <target_path>

Add optional overrides as needed:

--select <codes>
--ignore <codes>
--extend-select <codes>
--extend-ignore <codes>

Examples:

# Full project with defaults from pyproject.toml
ruff check

# One folder with defaults
python -m ruff check src/models

# Override to skip docs and TODO-like rules for this run
uv run ruff check src --extend-ignore D,TD

# Check only selected rules in a folder
ruff check src/data --select F,E9,I

Workflow

1. Baseline Analysis

Run <ruff_cmd> check with the selected scope and options.

Classify findings by type:

Autofixable safe.

Autofixable unsafe.

Not autofixable.

If no findings remain, stop.

2. Safe Autofix Pass

Run Ruff with --fix using the same scope/options.

Review resulting diff carefully for semantic correctness and style consistency.

Run <ruff_cmd> format on the same scope.

Re-run <ruff_cmd> check to refresh remaining findings.

3. Unsafe Autofix Pass

Run only if findings remain and allow_unsafe_fixes=true.

Run Ruff with --fix --unsafe-fixes using the same scope/options.

Review resulting diff carefully, prioritizing behavior-sensitive edits.

Run <ruff_cmd> format on the same scope.

Re-run <ruff_cmd> check.

4. Manual Remediation Pass

For remaining findings:

Fix directly in code when there is a clear, safe correction.

Keep edits minimal and local.

Run <ruff_cmd> format on the same scope.

Re-run <ruff_cmd> check.

5. Ambiguity Policy

If there are multiple valid solutions at any step, always ask the user before proceeding.
Do not choose silently between equivalent options.

6. Suppression Decision (# noqa)

Use suppression only when all conditions are true:

The rule conflicts with required behavior, public API, framework conventions, or readability goals.

Refactoring would be disproportionate to the value of the rule.

The suppression is narrow and specific (single line, explicit code when possible).

Guidelines:

Prefer # noqa: <RULE> over broad # noqa.

Add a brief reason comment for non-obvious suppressions.

If two or more valid outcomes exist, always ask the user which option to prefer.

7. Recursive Loop and Stop Criteria

Repeat steps 2 to 6 until one of these outcomes:

<ruff_cmd> check returns clean.

Remaining findings require architectural/product decisions.

Remaining findings are intentionally suppressed with documented rationale.

Repeated loop makes no progress.

Each loop iteration must include <ruff_cmd> format before the next <ruff_cmd> check.

When no progress is detected:

Summarize blocked rules and affected files.

Present valid options and trade-offs.

Ask the user to choose.

Quality Gates

Before declaring completion:

Ruff returns no unexpected findings for the chosen scope/options.

All autofix diffs are reviewed for correctness.

No suppression is added without explicit justification.

Any unsafe fix with possible behavioral impact is highlighted to the user.

Ruff formatting is executed in every iteration.

Output Contract

At the end of execution, report:

Scope and Ruff options used.

Number of iterations performed.

Summary of fixed findings.

List of manual fixes.

List of suppressions with rationale.

Remaining findings, if any, and required user decisions.

Suggested Prompt Starters

"Run ruff-recursive-fix on the whole repo with default config."

"Run ruff-recursive-fix only on src/models, ignore DOC rules."

"Run ruff-recursive-fix on tests with select F,E9,I and no unsafe fixes."

"Run ruff-recursive-fix on src/data and ask me before adding any noqa."