RepoDaily · 2026-06-21 · Dataset / Public directory

Matt Pocock's Skills: A Curated Skillset for Engineers Who Use AI Coding Agents

#3 Dataset / Public directory Shell +1,360 mattpocock/skills Open repository

A growing public directory of composable agent skills targeting the most common failure modes in AI-assisted software engineering — misalignment, verbosity, broken code, and architectural decay.

Repo typeDataset / Public directory
Best forEngineers using Claude Code, Codex, or similar agents who want lightweight, composable skills instead of heavyweight workflow frameworks
Risk levelLow — MIT-licensed text skills you can read, edit, and remove freely
Time to evaluate30 minutes to install and try one skill

Primary question: Do these small, composable skills actually fix the failure modes you hit with AI coding agents?

90/100

RepoDaily adoption score

RepoDaily rates this as 90/100 (strong) for adoption: evidence, installation path, production risk, differentiation, license clarity, and AI/agent fit are scored from the article sources and adoption notes.

Directional score from RepoDaily sources and adoption notes, not a benchmark.Risk: Low
100Evidence quality

5 source(s) across 5 source category/categories, plus a RepoDaily-specific evidence module when available.

88Installability

4 workflow step(s), 4 next-action step(s), and 2 command/install signal(s) were detected.

80Maintenance confidence

Trending momentum is +1,360 stars, with maintenance/release/issue signals counted when present.

97Production readiness

Risk is marked low, with 3 security note(s) and 3 explicit skip condition(s).

88Differentiation

3 opportunity lens item(s), 3 alternative(s), and 0 type-specific section(s) support differentiation.

82License clarity

License source or license wording is present.

84Agent / AI fit

6 AI/agent-related signal(s) were detected in the article text and metadata.

Project overview

mattpocock/skills is a public directory of agent skills that Matt Pocock uses daily for real engineering work. Unlike full-process frameworks such as GSD, BMAD, and Spec-Kit, these skills are intentionally small, easy to adapt, and composable. They work with any model and are designed to preserve developer control rather than take it away.

The repository is organized around four recurring failure modes in AI-assisted development: the agent not doing what you want, the agent being too verbose, the code not working, and the codebase becoming a ball of mud. Each skill targets one or more of these problems with a focused, repeatable practice.

Installation is handled through skills.sh. Running `npx skills@latest add mattpocock/skills` lets you pick the skills you want and which coding agents to install them on, followed by a `/setup-matt-pocock-skills` command that configures issue tracker integration and label conventions.

Problem it solves

  • Misalignment: you think the agent understands the task, then the output proves it didn't
  • Verbosity: agents dropped into unfamiliar projects over-explain because they lack a shared vocabulary
  • Non-working code: without tight feedback loops, agents produce code that fails silently or obviously
  • Architectural decay: agents accelerate coding so fast that codebases turn into a ball of mud

How it works

  1. Install via the skills.sh CLI: `npx skills@latest add mattpocock/skills`
  2. Select the skills you want and which agents to install them on, making sure to include `/setup-matt-pocock-skills`
  3. Run `/setup-matt-pocock-skills` to configure your issue tracker (GitHub, Linear, or local files), triage labels, and documentation output location
  4. Use user-invoked skills like `/grill-me` or `/grill-with-docs` to orchestrate sessions; model-invoked skills are triggered automatically when a task fits

How the Skills Are Organized

  • User-invoked skills (e.g. /grill-me, /grill-with-docs, /triage, /ask-matt) are triggered manually and orchestrate workflows
  • Model-invoked skills hold reusable discipline and can be reached for automatically by the agent
  • A user-invoked skill may invoke model-invoked skills but never another user-invoked one
  • Engineering skills cover daily code work; productivity skills cover non-code tasks

Flagship Skills to Try First

  • /grill-with-docs: a grilling session that also builds your project's domain model, sharpens terminology, and updates CONTEXT.md and ADRs inline
  • /tdd: encourages red-green-refactor and gives the agent guidance on what makes good and bad tests
  • /diagnosing-bugs: wraps best debugging practices into a simple loop
  • /improve-codebase-architecture: scans for deepening opportunities, presents a visual HTML report, then grills through your pick
  • /to-prd: quizzes you about which modules you're touching before creating a PRD

The Shared Language Technique

One of the standout ideas in this repo is the shared language document. Instead of letting the agent guess at jargon, you collaboratively build a CONTEXT.md that captures the ubiquitous vocabulary of your domain.

The README illustrates this with an example from Matt's course-video-manager repo, where a verbose sentence about lessons being 'made real' becomes the concise phrase 'materialization cascade.' This concision pays off session after session in fewer tokens and more navigable code.

Skill Architecture: Prompt Library With a Shared Command Surface

mattpocock/skills is a prompt-and-procedure directory for AI coding agents. The review target is not only the number of skills; it is whether `README.md`, `package.json`, shell scripts, and individual skill folders create a repeatable installation and update surface for developers.

Before adopting it as a team convention, run one flagship skill on a known codebase, record the exact command or agent invocation, and compare the output against your existing TypeScript or testing standards. The `LICENSE` and repository structure should be reviewed before copying skills into internal agent packs.

Who should pay attention?

Good fit if

  • You use Claude Code, Codex, or similar agents and keep hitting misalignment or verbosity issues
  • You want discipline and feedback loops without adopting a heavy process framework
  • You prefer skills you can read, edit, and compose over opaque automation
  • You value domain-driven design and want to bring it into AI-assisted workflows

Skip for now if

  • You don't use AI coding agents at all
  • You've already standardized on GSD, BMAD, or Spec-Kit and don't want a parallel system
  • You need a zero-setup experience — these skills require a one-time setup step and some customization

Risks and cautions

Low

MIT-licensed text skills that you install selectively and can remove at any time. No runtime dependency is introduced into your application code.

  • Skills are declarative prompt-like files, not libraries imported into your codebase
  • Setup is interactive and lets you choose exactly what to install
  • MIT license permits any use, modification, and redistribution
  • Model-agnostic design means no lock-in to a specific agent or vendor
  • Skills are stored as readable text files in the repo and can be inspected before installation
  • No telemetry or analytics are mentioned in the source pack
  • Setup asks where to save docs and which issue tracker to connect, giving you explicit control over integrations

Alternatives to compare

ApproachWhen to useTrade-off
GSD
When you want a full process framework that owns the entire workflow end to endOpen source
BMAD
When you want a structured methodology for agentic development with opinionated rolesOpen source
Spec-Kit
When you want specification-driven development with a heavier process layerOpen source

What this trend reveals

Fork and Specialize for Your Team

Because the skills are small and composable, teams can fork the directory and encode their own engineering standards — testing conventions, architecture rules, domain vocabulary — into agent-invokable skills.

Try forking one skill file and customizing the grilling questions for your project's domain.

Teach Agent Discipline to Junior Engineers

Skills like /tdd and /diagnosing-bugs encode decades of engineering best practices in a format the agent enforces. This can double as a teaching tool for developers learning red-green-refactor or systematic debugging.

Run /tdd on a sample feature with a junior engineer and compare the output quality to their usual workflow.

Build an Internal Skills Registry

The skills.sh installer pattern could be replicated inside an organization to distribute shared skills across teams, creating an internal marketplace of vetted agent behaviors.

Audit which skills your team would want to share and prototype an internal add command.

Best next action

Install the Core Skills and Try /grill-with-docs

The fastest way to understand the value of this repo is to install it, run setup, and then use the flagship grilling skill on a real change you're planning.

  1. Run `npx skills@latest add mattpocock/skills` and select /setup-matt-pocock-skills plus /grill-with-docs
  2. Run /setup-matt-pocock-skills to configure your issue tracker and documentation paths
  3. Pick a real, small change you want to make and run /grill-with-docs on it
  4. Compare the resulting CONTEXT.md and plan quality to your usual agent workflow

RepoDaily verdict

mattpocock/skills is a pragmatic, well-organized directory of composable agent skills that tackles the real failure modes of AI-assisted engineering. It earns its traction by offering discipline without dogma — small, readable, adaptable skills that work with any model and preserve developer control.

Sources