One engineer’s daily kit, not a process to surrender to
This repository is exactly what its tagline says: the agent skills Matt Pocock uses every day, straight from his .claude directory, for real engineering rather than vibe coding. The thesis is stated against the alternatives. Approaches like GSD, BMAD, and Spec-Kit try to help by owning the process, and in doing so take away your control and make bugs in the process hard to resolve. These skills go the other way: small, easy to adapt, composable, model-agnostic. You are meant to hack on them and make them your own, not hand over your workflow.
That positioning is the reason to choose it. If you want a complete methodology imposed, a framework is the right tool. If you want sharp individual fixes you stay in command of, this is built for that.
The grilling thesis at its core
The skills exist to fix concrete failure modes, and the headline one is misalignment: you think the agent understands what you want, then you see what it built and it clearly did not. The fix is a grilling session, getting the agent to interrogate you with detailed questions before it writes anything:
/grill-mefor non-code work, and/grill-with-docsfor engineering, which adds more context-gathering.
The author calls these his most popular skills, used every time before making a change, and the logic is sound: the cheapest place to catch a misunderstanding is before code exists.
Beyond grilling, the kit handles the other recurring frictions of real work. A /triage skill sorts issues using the labels you actually apply, so it slots into your existing tracker rather than imposing its own taxonomy. A /handoff skill captures the state of a session so you, or another agent, can resume cleanly instead of re-deriving context. And because every skill is small, single-purpose Shell that works with any model, you compose the few you need rather than adopting a monolith, and you edit them when your workflow differs. That composability is the practical difference from a framework: you take the two or three skills that fix your actual pain and leave the rest.
Install
A single command pulls them in:
npx skills@latest add mattpocock/skillsPick the skills you want and the agents to install them on, and make sure you select /setup-matt-pocock-skills. Then run that setup skill in your agent; it asks which issue tracker you use (GitHub, Linear, or local files), what labels you apply when triaging (the /triage skill uses them), and where to save any docs created. After that you are ready.
The gotchas, and they are honestly model-dependent
This is where the issue tracker earns its place, because the failures cluster around the grilling skill behaving differently across models:
- Over-grilling. One thread reports Codex asking 200 questions after a grill, the alignment skill taken to an extreme. The right amount of interrogation is itself model-dependent.
- Under-grilling. Another reports GPT-5.5 often jumping straight to implementation after
/grill-me, ignoring the intended pause. So the skill’s effect varies with the model you run. - Portability bugs. The
handoffskill’smktemp -t handoff-XXXXXX.mddoes not produce the intended filename on macOS because BSDmktempdiffers from GNU, a classic shell-portability trap.
These are the honest seams of small shell skills that run across many models and platforms. With 68 open issues as of 2026-06, the surface is small and the fixes are tractable. Because the skills are meant to be adapted, hitting one of these is also an invitation to edit the skill, which is the whole philosophy.
mattpocock/skills versus the alternatives
| mattpocock/skills | superpowers | anthropics/skills | |
|---|---|---|---|
| Stars | 123,769 | 222,999 | 148,807 |
| Stance | small, you keep control | full methodology, owns process | reference and spec |
| Adapt | designed for hacking | opinionated workflow | fork from examples |
| Best for | real-engineer daily habits | imposing discipline | learning the format |
Counts are from GitHub as of June 2026. superpowers is the process-owning methodology this repo explicitly contrasts itself with: powerful, but it takes the wheel. anthropics/skills is the neutral reference. mattpocock/skills is the practitioner’s middle path: real, composable fixes you keep editing, with no framework deciding how you work.
Related
For the heavier methodology this kit positions against, see superpowers; for the official reference, anthropics/skills. Manage skills across tools with cc-switch. For what else is climbing, see LLM tooling, the daily digest, and the weekly report.
FAQ
What problem do these skills solve? Concrete agent failure modes, led by misalignment. The /grill-me and /grill-with-docs skills make the agent interrogate you before writing code.
How do I install them? npx skills@latest add mattpocock/skills, select /setup-matt-pocock-skills, then run it to configure your issue tracker, labels, and docs location.
Why does grilling behave differently for me? It is model-dependent. Reports show Codex over-questioning and GPT-5.5 skipping the pause, so tune it for your model and edit the skill if needed.
How is this different from a framework like Superpowers? It is deliberately small and composable and leaves you in control, rather than owning your whole process the way GSD, BMAD, or Spec-Kit do.