4 free GitHub repos for Claude (and the one I won't install)

A ruleset Claude Code loads every turn. Before writing code Claude has to ask: does this already exist, is it built in, can it be one line? In our test (a wedding countdown, 3 runs each) plain Claude averaged 68 lines and Ponytail 26.

1. Send the first command as its own prompt in Claude Code.
2. Send the second command as a separate prompt.
3. Start a new session. Type /ponytail to check it's on.

Tip: Biggest wins on UI widgets and helper files. On code that's already minimal it changes almost nothing.

Official docs →

Turns a project into a knowledge graph Claude can query instead of re-reading files. You get graph.html (clickable map), GRAPH_REPORT.md (key concepts and surprising connections) and graph.json.

1. Install the CLI. The package name is graphifyy, with two y's; other graphify* packages are not affiliated.
2. Register the skill with your assistant.
3. In Claude Code, run /graphify . inside your project.

Tip: Mapping code runs locally. Extracting PDFs and documents uses an AI API key, which may cost money.

Official docs →

25 skills plus slash commands that walk Claude through the full lifecycle: /spec, /plan, /build, /test, /review, /code-simplify and /ship. Each one encodes the quality gates senior engineers use.

1. Install all 25 skills with one command, or add --list first to browse them.
2. Or install a single skill, for example code review.
3. Start with /spec on your next feature.

Tip: 25 skills add context to every session. If you already have review or testing skills, install just the ones you're missing.

Official docs →

A local AI gateway: one endpoint that routes to 350+ providers and falls back to free tiers when one runs out, so you rarely hit a limit. The catch: to use your own accounts you put your provider API keys and logins into it, so one third-party app holds all of them. It also ships a MITM mode that installs its own certificate authority to read encrypted traffic, plus TLS-fingerprint stealth. Some of that pushes against providers' terms of service and could get accounts banned.

1. Read the source and the security docs before trusting it with any key.
2. If you test it, use throwaway keys with low spending limits, never your main ones.
3. Don't install its certificate authority (MITM mode) on a machine you use for anything sensitive.

Tip: It's MIT-licensed and popular, and I'm not saying it's malware. My rule: anything that sees all my keys and my encrypted traffic needs a much higher bar than a star count.

Official docs →