gierd

Gierd Skills

[!WARNING] This repository is deprecated. It has been replaced by gierd-inc/g-hive. No further updates will be made here. Please migrate to g-hive.

Plugin name: gierd. Commands: /gierd:build, /gierd:prd-to-spec, /gierd:status, etc.

Agent skills for Gierd's Ruby on Rails engineering workflow — daily-driver process tools, the Gierd Dev Skills PRD-to-shipped pipeline, and a complete set of Rails-domain skills.

Forked from mattpocock/skills and ryenski/agency-plugin. Process skills (grill, triage, to-prd, to-issues, diagnose, tdd, improve-architecture) come from Matt Pocock's set; Rails skills and the Gierd Dev Skills workflow come from the agency-plugin. Both are credited inline in their respective SKILL.md files where retained verbatim.

What's in this repo

These skills are designed to be small, easy to adapt, and composable. They work with any model. Hack around with them. Make them your own.

The skills cover six broad areas:

1. Process — grill the agent into alignment, triage incoming bugs, break PRDs into issues, debug systematically.
2. Rails domain knowledge — one focused skill per Rails layer (models, controllers, views, jobs, mailers, migrations, security, performance, testing, Hotwire, etc.), targeted at Rails 8.1+. Each is loaded on-demand based on what the agent is touching.
3. OOP extraction patterns — value objects, null objects, concerns, presenters, query objects, form objects, service objects, policy objects, and (rarely) repository pattern. Loaded when fat models stop being enough.
4. Rails antipatterns — named smells (fat-controller, callback-hell, service-object-soup, etc.) drawn from Rails Antipatterns (Pytel & Saleh) and the 37signals tradition. Loaded when reviewing or refactoring code that exhibits a smell.
5. The Gierd Dev Skills workflow — PRD → Spec → Issues → Code → draft PR, with file-based subagent contracts and resumable per-feature state under .dev-skills/.
6. Productivity — grill-me, caveman mode, write-a-skill.

Installation

This is a private Claude Code plugin. Install it through Claude Code's plugin marketplace system.

From GitHub

Add the marketplace and install with the claude plugin CLI:

claude plugin marketplace add gierd-inc/dev-skills
claude plugin install gierd@gierd

You'll need GitHub access to gierd-inc/dev-skills (SSH key or HTTPS credentials configured for git).

From a local directory

If you've already cloned the repo somewhere, point Claude Code at the path:

claude plugin marketplace add ~/Projects/dev-skills
claude plugin install gierd@gierd

After installing

Run /setup-gierd-skills in your agent. It will:

• Ask which issue tracker you use (Linear by default for Gierd repos, with GitHub / GitLab / local-markdown as alternatives)
• Confirm the triage label vocabulary
• Confirm whether the repo is single- or multi-context (multi-context is default for repos with CONTEXT-MAP.md)

Development

If you're actively editing the skills, install via the local directory route above. Claude Code reads from your working tree, so edits to any SKILL.md (or command, agent, etc.) are picked up on the next agent session — no reinstall step:

claude plugin marketplace add ~/Projects/dev-skills
claude plugin install gierd@gierd

For lightweight iteration on just the skills (without registering as a full plugin), use the bundled symlink scripts:

./scripts/link-skills.sh    # symlink every skill into ~/.claude/skills/
./scripts/unlink-skills.sh  # remove the symlinks created above

link-skills.sh symlinks every SKILL.md directory under skills/ (excluding deprecated/) into ~/.claude/skills/. unlink-skills.sh removes only the symlinks pointing back into this repo, leaving any unrelated entries in ~/.claude/skills/ untouched. Useful when you only need the skills loaded and don't care about the /gierd:* slash commands or Gierd Dev Skills subagents.

Slash-command wrappers

Skills auto-load by description match, but they don't appear in the / picker. To make every skill invokable as a slash command, run:

./scripts/generate-skill-commands.sh

This walks skills/**/SKILL.md and writes a thin wrapper into commands/ for each one — so skills/oop/oop/SKILL.md becomes /gierd:oop, skills/gierd/deploy-to-dev/SKILL.md becomes /gierd:deploy-to-dev, etc. Re-run it after adding, renaming, or removing skills.

The script is idempotent: wrappers carry an HTML comment marker, and only marked files are touched. Hand-written commands (build.md, prd-to-spec.md, etc.) are preserved.

Why these skills exist