bbcode-autofix

name	bbcode:autofix
description	Autofix common code-quality issues by running a pipeline of focused sub-agents (one-class-per-file, etc.). Use when the user wants to clean up code smells, normalize file structure, or run autofix on a set of files.
argument-hint	[files-or-glob | description of target]

Autofix — Run Common Code-Quality Fixers

Apply a pipeline of focused sub-agents to detect and fix common code-quality issues in a target set of files. Each sub-agent in agents/ owns one specific issue class.

Arguments

$ARGUMENTS — Optional. Either a list of files / globs, or a natural-language description of which files to target (e.g. src/**/*.ts, the auth module, everything I changed since main). If omitted, target all uncommitted files in the working tree.

Step 1 — Determine Target Files

Two branches:

Branch A: No argument

Inspect uncommitted files in the working tree:

git status --porcelain

Build the target list from:

• Modified files (M, M)
• Added files (A, ??)
• Renamed files (R — use the new path)

Skip:

• Deleted files
• Files outside the repo
• Anything matching .gitignore patterns the working tree happens to surface

Branch B: Argument provided

Interpret $ARGUMENTS as the targeting instruction:

• If it looks like a file path or glob → expand it directly (use git ls-files or globbing)
• If it looks like a natural-language description → resolve it to a concrete file list (ask the user to confirm if ambiguous)

Step 2 — Discover Sub-Agents

List every *.md file in agents/ (relative to this skill). Each file is a sub-agent specification: its body is the prompt, its filename is its identifier.

Each sub-agent file follows this shape:

• A Purpose section — what issue it fixes
• A Detection section — how it identifies affected files
• A Fix section — what changes it makes
• A Constraints section — what it must not touch

Step 3 — Run the Sub-Agent Loop

For each sub-agent file in agents/:

1. Spawn a sub-agent via the Agent tool with subagent_type: "general-purpose".
2. Prompt it with:
   • The full content of the sub-agent's .md file as instructions
   • The resolved target file list from Step 1
   • An explicit instruction to apply changes directly (Edit/Write), not just report
3. Wait for the sub-agent to complete and capture its summary.
4. Report back to the user: which files the sub-agent touched and a one-line summary of changes.

Run sub-agents sequentially, not in parallel — later fixers may depend on the structure produced by earlier ones (e.g. once single-class-per-file has split a file, naming/import fixers can run on the result).

Step 4 — Final Summary

After all sub-agents complete, print:

## Autofix Summary

Target: <N files>
Sub-agents run: <N>

### <sub-agent-name>
- Files touched: <list>
- Changes: <one-line summary>

### <sub-agent-name>
- ...

Run tests / typecheck before committing.

Rules

• Never commit — autofix only edits the working tree. The user reviews and commits.
• Never run destructive git commands (reset, clean, checkout --) to "reset" between sub-agents. The pipeline is additive.
• If a sub-agent fails or reports it cannot safely fix something, surface that in the summary and continue with the next sub-agent — do not abort the whole pipeline.
• Always use repo-root-relative paths in reports.
• If the target file list is empty, say so and exit; do not invent files.