Skills for Real Engineers. Straight from my .claude directory.
Created 2026-02-03 — roughly 3.5 months old at time of appraisal. No formal releases have been cut, which means no semantic versioning or changelog discipline. That said, the installer (npx skills@latest) implies a published npm package with implicit versioning. 25 open issues is low relative to 74,970 stars, suggesting either strong triage or low defect surface area (the latter is plausible since most content is prose skill files). Commit cadence appears strong — the last commit is the same day as this appraisal. Functionally stable for its purpose, but the absence of tagged releases prevents a higher rating.
README is among the better-structured project READMEs in the vault. It opens with a clear value proposition, frames four named failure modes with engineering book citations (Pragmatic Programmer, DDD, XP Explained, Philosophy of Software Design), maps each to a specific skill, and provides a 30-second quickstart. Each skill has its own SKILL.md linked from a reference section. The setup wizard pattern (/setup-matt-pocock-skills) means first-run experience is self-guided. The philosophy section doubles as onboarding documentation. Beyond the README, individual SKILL.md files provide per-skill context.
The primary artifact is prompt/skill text files, not executable code — traditional quality signals (tests, CI, linting) don't directly apply. The Shell component appears to be the installer glue. No dependency manifest is available, and no test presence or CI badges are visible in the README. The prompt engineering quality, however, is evidently high: skills are deliberately small and composable, each scoped to a single failure mode, which reflects good separation of concerns. The distribution mechanism (npx skills@latest) implies an npm package with some infrastructure behind it. Quality ceiling is capped at 3 due to absence of verifiable code-quality signals.
Last commit matches today's date (2026-05-12), and the repo was created only 3.5 months ago with presumably continuous activity to accumulate nearly 75k stars. Author Matt Pocock is a well-known practitioner (Total TypeScript) with a demonstrated track record of active project maintenance. The 25-issue count is responsibly low for the star count. Newsletter subscription (~60,000 mentioned in README) provides a direct feedback loop that incentivizes continued maintenance.
74,970 stars and 6,468 forks in ~3.5 months is an exceptional trajectory — this is a breakout hit by any standard. The skills.sh badge signals integration into a broader ecosystem beyond GitHub. A newsletter with ~60,000 subscribers provides a distribution channel that amplifies adoption. Forks-to-stars ratio (~8.6%) is notably high, indicating practitioners are actively copying and adapting the skills rather than merely starring — a strong signal of practical utility.
Overall: 4.1/5
Category: AI Coding Agent Skills
Known alternatives in vault: jkomoros--prompt-garden (LLM Prompt Composition, 1.6/5) — loose conceptual overlap only; no direct equivalent exists in the vault.
Differentiation: prompt-garden is about static prompt composition and storage; this repo provides opinionated, engineering-philosophy-grounded workflow skills designed for interactive coding agents (Claude Code, Codex, etc.). The key distinctions are: (1) skills are scoped to named failure modes rather than generic prompts, (2) composability is first-class — skills chain (grill-with-docs → to-prd → tdd), (3) a shared installer ecosystem (npx skills@latest, skills.sh) provides structured distribution, and (4) the CONTEXT.md / ubiquitous language pattern is a novel knowledge-capture primitive not present elsewhere in the vault. prompt-garden offers nothing comparable in terms of engineering workflow structure or agent-specific integration.
Gap or crowd: Clear gap. No repo in the vault addresses curated, opinionated coding-agent skill libraries. This is the sole representative of a category that is emerging rapidly alongside agentic coding tools.
Score: 5/5
Harvestable: /grill-me and /grill-with-docs (structured pre-task alignment questioning); /tdd (red-green-refactor discipline for agents); /diagnose (reproduce→minimise→hypothesise→instrument→fix→regression-test debugging loop); /to-prd (module-aware PRD generation); /zoom-out (system-level code explanation); /improve-codebase-architecture (architecture rescue); CONTEXT.md ubiquitous-language pattern; ADR (Architectural Decision Record) generation integration.
Integration path: All skill files can be copied directly into any agent's instruction directory (.claude/, .cursor/rules/, system prompts) via the npx skills@latest installer or manual extraction. The CONTEXT.md pattern is immediately adoptable as a vault-level primitive — a per-project shared-language document that reduces token overhead and improves naming consistency across the PAI system. The /grill-me pattern maps directly onto PAI intake workflows for new tasks or knowledge capture sessions.
Overlap with existing: jkomoros--prompt-garden covers prompt storage but with no workflow structure and a 1.6/5 standalone score; overlap is nominal. No other vault repo provides coding-agent skill primitives.
Adoption cost: Trivial. The npx installer handles placement. Manual extraction of any individual SKILL.md is a copy-paste operation. No build step, no runtime dependency, no infrastructure required. The only effort is selecting which skills to activate and running /setup-matt-pocock-skills once per project.
This repo is the highest-starred item encountered in this appraisal run by a wide margin (~75k vs the next at ~4.8k for influxdata--influxdb territory, though that's a much older project). The star trajectory in under four months is anomalous and warrants a note: part of this may reflect Matt Pocock's existing audience and the broader cultural moment around agentic coding in early 2026. However, the fork rate (~8.6%) suggests genuine adoption rather than passive interest. The skills.sh ecosystem (badge present, npm package) hints at an emerging standard for skill distribution, analogous to how Homebrew formulas or npm packages work — worth monitoring as a potential infrastructure dependency. The PAI system should treat this repo not just as content to import but as a pattern source: the failure-mode → named skill → composable chain structure is a design vocabulary worth applying to PAI's own skill authoring.