aihero-git-guardrails-claude-code

Agent Skills — fork (Claude Code plugin packaging)

⚡ Quick install: pick any skill (or several)

Add the marketplace once, then install whichever skills you want — one at a time, no all-or-nothing:

/plugin marketplace add VariousForks/skills-by-aihero-dev-mattpocock
/plugin install aihero-tdd@variousforks-mattpocock-skills

Then /reload-plugins. The skill is now invocable as aihero-<name>:<name> (e.g. aihero-tdd:tdd). The shorthand resolves to the fork's default branch (main-gw) automatically — no branch specifier needed. If you'd rather pin to an exact branch/tag, use the full-URL form: /plugin marketplace add https://github.com/VariousForks/skills-by-aihero-dev-mattpocock.git#main-gw.

Available plugins (each can be installed independently):

• aihero-tdd@variousforks-mattpocock-skills — test-driven development loop
• aihero-write-a-prd@variousforks-mattpocock-skills — PRD interview
• aihero-prd-to-plan@variousforks-mattpocock-skills — PRD → multi-phase plan
• aihero-prd-to-issues@variousforks-mattpocock-skills — PRD → GitHub issues
• aihero-grill-me@variousforks-mattpocock-skills — Socratic plan review
• aihero-git-guardrails-claude-code@variousforks-mattpocock-skills — block dangerous git
• aihero-improve-codebase-architecture@variousforks-mattpocock-skills — architecture review
• aihero-triage-issue@variousforks-mattpocock-skills — bug triage with TDD-fix plan

Each of the 8 plugins was packaged on its own feat/marketplace-with-<skill>-as-installable-plugin branch (one PR's worth of changes per skill, branched from main-upstream) and consolidated here on main-gw. Independent issues and PRs against this fork are welcome — see CONTRIBUTING.md for the contribution culture (Why?-driven descriptions, example-based testing, AI co-authorship transparency). CONTRIBUTING-PLUGINS.md is the recipe for packaging another skill from this repo.

What you get after install

Once aihero-tdd:tdd is loaded, asking Claude for TDD-driven work changes its behavior in three specific, observable ways:

• Interview before code. Claude pauses and asks for the public interface, the first behavior to drive, and a priority list — instead of jumping straight to implementation. This is the skill's anti-pattern guardrail against bulk-test-writing followed by bulk-implementation.
• Strict vertical slicing. One test → one minimal implementation → run → repeat. No multi-test bursts, no implementation that runs ahead of the test that justifies it.
• Explicit RED / GREEN narration. Claude literally announces RED confirmed (…) after the failing run and GREEN. Next behavior: … after the passing run, giving you a clean audit trail when reviewing the session.

A full reproducible verification — empty Go module + Greet(name string) string driven from zero through two red-green cycles — is captured in TDD-SKILL-IN-ACTION.md. Useful both as a sanity check after you install (does the skill actually trigger?) and as a transcript-style example of what the loop looks like in practice.

If after install Claude doesn't enter the interview / RED-GREEN flow on a TDD prompt, the most common cause is that /reload-plugins wasn't run after install, or the prompt didn't surface the trigger words from the skill description (tdd, red-green-refactor, test-first, integration tests). Trying again with one of those words usually fixes it.

This is a fork of mattpocock/skills. All skill content (the SKILL.md files and their companion guides) is Matt Pocock's / AI Hero's work. This fork adds packaging only: the tdd skill is wired up as a Claude Code plugin (.claude-plugin/plugin.json), and a marketplace catalog (.claude-plugin/marketplace.json) at the repo root makes selective installation possible. Upstream feature request and full PR description: mattpocock/skills#138.

The default branch here is main-gw (Greg's customised trunk). The pristine upstream main is preserved as main-upstream for clean future syncs / PRs.

🤝 Greg (human) and Claude (AI) co-built the packaging — with much respect for Matt's TDD work.

A collection of agent skills that extend capabilities across planning, development, and tooling.

Planning & Design

These skills help you think through problems before writing code.