Smart-git-commit AGENTS.md

Smart Git Commit — Agent Skills Manifest

Skill Directory Structure

skills/
  smart-git-commit/       # kebab-case directory name
    SKILL.md              # Required: skill definition and phases
    scripts/              # Executable bash scripts
    references/           # Supporting docs, loaded on demand
    templates/            # Drop-in config files for projects
    tests/                # Test fixtures and scenarios

Naming Conventions

• Skill directories: kebab-case
• SKILL.md: always uppercase, always this exact filename
• Scripts: kebab-case.sh (bash) or kebab-case.mjs (Node)
• References: kebab-case.md
• Templates: dotfiles and .example suffixes (.gitmessage, CLAUDE.md.example)

Script Requirements

• Shebang: #!/usr/bin/env bash
• Error handling: set -euo pipefail for strict mode
• Progress/logs → stderr
• Machine-readable results → stdout as JSON
• Cleanup traps for temp files
• Exit 0 = success, exit 1 = failure (blocks caller)
• Color: Use ANSI color codes (RED for errors, GREEN for success, YELLOW for warnings)

Output JSON Schema

All scripts output JSON to stdout:

{
  "status": "pass" | "fail" | "not_found",
  "findings": [
    {
      "file": "path/to/file.ext",
      "line": 42,
      "pattern": "matched pattern description"
    }
  ],
  "message": "Human-readable summary"
}

SKILL.md Rules

• Under 500 lines
• YAML frontmatter with name + description (description must be "pushy")
• Progressive disclosure: SKILL.md is entry point, details in references/
• Phases numbered, each references the right script or reference file

Progressive Disclosure Pattern

1. Level 1 — Frontmatter (always in context, ~100 words)
2. Level 2 — SKILL.md body (in context when skill triggers, <500 lines)
3. Level 3 — references/ (loaded on demand, unlimited depth)
4. Level 4 — scripts/ (executed, only output enters context)
5. Level 5 — templates/ (drop-in configs, user copies and customizes)