Skills for Real Engineers. Straight from my .claude directory.
npx skills@latest) with a skills.sh registry.| Rating | Summary | |
|---|---|---|
| Quality | solid (17/24) | Exceptionally well-documented engineering skill set with extraordinary adoption signals; held back only by the absence of tagged releases and any formal test infrastructure. |
| PAI Relevance | integrate (0.63) | Language-agnostic markdown skills that port directly into PAI's skill framework, filling genuine gaps in domain-language building and structured TDD workflow for coding agents. |
Prior landscape entry showed poor (4.1/24); fresh appraisal with 7/10 evidence density and current star count yields solid (17/24), flipping the verdict from note to integrate. The prior score likely reflects an early-stage snapshot of a fast-moving repo.
17/24 — maintained / well-documented / solid
Failed:
latest_release: none.github/workflows reference foundPassed:
archived: falseFailed:
Passed:
npx skills@latest add mattpocock/skills install commandFailed:
Passed:
| Dimension | Score | Assessment |
|---|---|---|
| Harvest Value | 1 | The /grill-with-docs pattern — building a shared CONTEXT.md and ADRs during requirement gathering — is novel within PAI's current skill set and worth studying for improving the Interview and ISA skills; the diagnose loop and architecture-review cadence are also useful reference points. |
| Integration Readiness | 2 | Skills are plain SKILL.md markdown files — fully language-agnostic and directly portable; the installer runs via npx/bunx which PAI's Bun-based stack handles natively without adapter code. |
| Overlap Risk | 1 | Partial overlap: /diagnose overlaps with PAI's RootCauseAnalysis skill, /grill-me partially overlaps with Interview, and /to-prd partially overlaps with ISA — but no single skill is a complete duplicate of an existing PAI capability. |
| Gap Fill | 1 | PAI has no dedicated domain-vocabulary-building skill (CONTEXT.md + ADR generation during planning), no structured TDD-enforcement skill for coding agents, and no periodic architecture-review skill — all three are addressed here. |
Composite: 0.63
Capture-to-Knowledge Pipeline is in design phase with manual validation still running: Run npx skills@latest in the pipeline's working directory to install the skills, then invoke the alignment grilling skill at the start of each new implementation session before any code is written — agent-proposed implementations get interrogated against design intent before they diverge, not after a session's worth of output has to be unwound.
Aphoria (TypeScript, Bun, Stage 3A complete) has domain-specific vocabulary — RiTa primitives, generation stages, aphorism structures — that agents consistently drift away from across sessions: Invoke the domain-language skill once to extract and codify that vocabulary into a persistent agent context file — subsequent Claude Code sessions use the project's own terminology rather than reinventing it, keeping generated code consistent with the existing codebase's conceptual model.
Conservancy (SvelteKit, Cloudflare Pages) has a mature editorial pipeline that will eventually need new feature work: Install skills at the repo root now while the pipeline is stable, so the TDD loop and architecture review skills are in place before the next feature cycle begins — new functionality gets spec'd, tested, and reviewed through structured commands rather than open-ended prompting against an unfamiliar codebase.
Category: Developer Tools & CLIs
In this category: casey--just
Standing: Where casey--just provides a stable, discoverable subprocess interface for running project recipes, mattpocock/skills provides the upstream cognitive layer — structured slash commands that encode how an agent should think, align, and plan before writing any code.
Density: 7/10 — README (first 8KB) available and substantive; star, fork, issue, commit, and license metadata available; no dependency manifest; no CI configuration visible; no release history; individual SKILL.md files not directly inspected beyond README excerpts and the Reference section.
The 108k-star count in roughly four months is the strongest virality signal in the current landscape — comparable to or exceeding most established developer tools. The repo is philosophically adjacent to humanlayer--12-factor-agents and bmad-code-org--BMAD-METHOD but deliberately avoids owning the full development process, a design distinction the README calls out explicitly as a feature. The overlap cluster with forrestchang--andrej-karpathy-skills is worth watching: both encode engineering best practices as agent-readable markdown, and the two skill sets are complementary rather than duplicative.