collect-commit-info

This skill analyzes staged git files and returns commit specifications as JSON. It:

1. Collects staged file metadata via script
2. Reads diff content to understand changes
3. Groups files by category (deps, ci, config, source, test, docs)
4. Generates conventional commit messages for each group
5. Returns JSON specs for the caller to execute

Arguments

| Arg | Default | Description | |-----|---------|-------------| | --lang | context | Message language (conversation context → system locale → en) |

Output Contract

This skill ALWAYS returns JSON. The caller executes git commands based on this output.

Success:

{
  "commits": [
    {"message": "type(scope): subject\n\nBody", "files": ["file1", "file2"]}
  ],
  "summary": {"total_commits": 1, "total_files": 2}
}

Error:

{"error": "description", "error_code": "CODE"}

Workflow

Step 1: Collect file info

${CLAUDE_PLUGIN_ROOT}/skills/collect-commit-info/scripts/collect-info.sh --lang <code>

Script returns JSON with temp_dir, files array (with categories), and paths.diff_content.

On error JSON → return it immediately and stop.

Step 2: Read diff content

Read the file at paths.diff_content to understand what changed.

Step 3: Group files by category

Group files from the files array. Commit order:

1. deps - Dependencies (package.json, lock files)
2. ci - CI/CD (.github/*)
3. config - Configuration (*.yml, *.toml, rc files)
4. source - Source code
5. test - Tests (_test., .test., .spec.)
6. docs - Documentation (.md, docs/)

Step 4: Generate commit messages

For each group, create a conventional commit message:

type(scope): imperative subject (max 50 chars)

Optional body explaining what and why.

Type by category: | Category | Type | |----------|------| | deps | chore | | ci | ci | | config | chore | | source | feat/fix/refactor (analyze diff) | | test | test | | docs | docs |

For source files, analyze diff to determine type:

• New functionality → feat
• Bug fixes, error handling → fix
• Restructuring, cleanup → refactor

Scope: Derive from paths (e.g., src/auth/* → auth)

Language: Use lang.effective from script output. If conversation context indicates a preferred language, use that instead.

Step 5: Cleanup and return

${CLAUDE_PLUGIN_ROOT}/skills/collect-commit-info/scripts/cleanup.sh <temp_dir>

Return the JSON commit specs.