📦 About the Repository

Organized fork of the skill set published by Matt Pocock — packaged as a Claude Code plugin.

Understanding the provenance helps you trust the conventions and track upstream.

Curated fork, upstream, Claude Code plugin, authorship.

Each skill is small, chainable, and adaptable—the opposite of monolithic frameworks.

These 3 values explain why skills outperform giant agents in real projects.

Single responsibility, composition, fork-friendly.

engineering = day-to-day coding; productivity = non-coding workflows; misc = maintained but rare.

Knowing which bucket to look in helps you avoid wasting time digging through the repo.

Categorization, discovery, intended use.

Direct comparison: GSD/BMAD/Spec-Kit are closed frameworks; mp-skill is a loose collection of skills.

Choosing the right tool for each problem — not everything is a skill.

Framework vs. toolkit, lock-in, granularity.

TypeScript-heavy full-stack developers, solo builders, small teams that use Claude Code/Codex every day.

Knowing the target audience helps calibrate expectations and contribute well.

Persona, real use cases, community.

Permissive license + PR-based contribution flow — open to new skills and improvements.

Know what you can redistribute, adapt, and send back without friction.

License, PR, CODEOWNERS, attribution.

Bucket with coding skills: grill-me, tdd, diagnose, improve-codebase-architecture, etc.

It's where 80% of daily use happens — it's worth getting to know each one.

Core skills, gateway, slash commands.

Bucket with process skills: handoff, statusline, audit, planning — outside the editor.

Process skills save the most time and context between sessions.

Workflow, handoff, session.

Skills that are maintained but used sporadically — they don’t deserve prominence, but they don’t deserve to be deleted either.

Find the right skill at the right time. It’s worth reviewing misc quarterly.

Long tail, minimal maintenance.

personal = owner's setup; in-progress = drafts; deprecated = retired — none appear in the README/plugin.json.

Understanding why skills disappear from the index avoids confusion when contributing.

Lifecycle, draft, promotion, deprecation.

docs/ stores references; scripts/ contains utilities; CLAUDE.md provides organization rules for the repo itself.

Without understanding CLAUDE.md, you break conventions when adding a new skill.

Project, automation, and documentation rules.

plugin.json lists published skills; README lists each skill with a link to SKILL.md.

Every new skill needs to be added to both—or it will be invisible in the plugin.

Manifest, index, publication gates.

Have Claude Code or Codex installed, with a compatible plan and a working shell.

Without the right runtime, skills don’t even appear — debugging that later is frustrating.

Runtime, plan, shell, plugin host.

Single command that downloads and registers the plugin in Claude Code/Codex.

It's the official way — other installation methods break during updates.

npx, registry, marketplace, plugin.

The interactive wizard lets you choose which skills to enable—starting with engineering skills is recommended.

Too many skills for an agent = noise. Initial curation is worth its weight in gold.

Curation, scope, selective activation.

Slash command that creates an initial CONTEXT.md, configures defaults, and validates triggers.

Skips a lot of manual setup and ensures skills trigger at the right times.

Bootstrap, defaults, triggers.

mp-skill GitHub Issues with labels (bug, skill-request, docs) — public demand queue.

Using the right label speeds up the response and prevents duplicate issues.

Triage, labels, priority, community.

Checklist: list the skills in the agent, run /grill-me in a dummy repo, and check the logs.

Validating early prevents you from discovering that installation failed only when you try to use it.

Smoke test, dry run, logs.