fastc

Zero-config CLI tool that analyses your git diff with an LLM and creates atomic, conventional commits.

$ fastc
analysing all changes (12 files)...
  using two-phase approach for large diff
  phase 1: grouping files...
  phase 2: generating message for group 1/2...
  phase 2: generating message for group 2/2...
  feat(auth): add JWT token validation
  refactor(db): extract connection pool config
  chore: update lockfiles
pushed

Features

• Atomic commits - Groups related changes into logical commits
• Conventional commit format - feat, fix, refactor, chore, docs, style, test, perf
• Smart diff handling - Two-phase approach for large diffs (15+ files)
• Lockfile separation - Commits lockfiles separately with chore: update lockfiles
• Diff compression - Reduces token usage for large diffs while preserving context
• Retry with backoff - Automatic retry on API rate limits or transient errors
• Rename detection - Properly handles renamed files

Install

Requires uv.

# Clone and symlink (use absolute path)
git clone https://github.com/danwt/demo-fast-commit.git
cd demo-fast-commit && ln -s "$(pwd)/fc" ~/.local/bin/fastc && cd ..

# Create config
mkdir -p ~/.config/fast-commit
cat > ~/.config/fast-commit/.env << 'EOF'
OPENROUTER_API_KEY=sk-or-v1-your-key-here
MODEL=google/gemini-2.0-flash-001
EOF

Get an API key from openrouter.ai/keys.

Usage

cd your-git-repo
fastc                # analyse, commit, and push
fastc --dry-run      # preview commits without executing
fastc --no-push      # commit but don't push
fastc --no-verify    # bypass pre-commit hooks

The tool will:

• Use staged changes if any exist
• Otherwise stage and commit all changes
• Push after committing (unless --no-push)

Config

~/.config/fast-commit/.env:

Variable	Description
OPENROUTER_API_KEY	Your OpenRouter API key (required)
MODEL	Model to use (required). Any model on OpenRouter works.
STRUCTURED_OUTPUT	Set to false to disable JSON mode (default: true)

Recommended models

Model	Speed	Quality	Cost
google/gemini-2.0-flash-001	Fast	Good	Low
anthropic/claude-3.5-sonnet	Medium	Excellent	Medium
openai/gpt-4o-mini	Fast	Good	Low

How it works

1. Detect changes - Uses staged changes if any, otherwise stages everything
2. Exclude noise - Filters out lockfiles and generated files from analysis
3. Analyse diff - For small diffs, sends directly to LLM. For large diffs (15+ files), uses two-phase approach:
   • Phase 1: Send file summary to group files by logical change
   • Phase 2: Generate commit message for each group
4. Create commits - Stages files per group and commits with generated message
5. Commit lockfiles - Any lockfiles get a separate chore: update lockfiles commit
6. Push - Pushes all commits to remote

Excluded from analysis

These files are excluded from LLM analysis (but still committed):

• Lockfiles: pnpm-lock.yaml, package-lock.json, yarn.lock, bun.lockb, Cargo.lock, poetry.lock, Pipfile.lock, go.sum, composer.lock, Gemfile.lock
• Generated: *.min.js, *.min.css, *.map

Claude Code integration

Copy SKILL.md to your .claude/skills/ directory:

mkdir -p ~/.claude/skills
cp SKILL.md ~/.claude/skills/fastc.md

Then invoke with /fastc in Claude Code.

License

MIT