sanyuan0704/code-review-expert
5 skills1.8k installs18.1k starsGitHub
Install
npx skills add https://github.com/sanyuan0704/code-review-expertSkills in this repo
1Skill ForgeSkill-forge is a meta agent skill that teaches you how to create high-quality, production-grade skills for Claude Code instead of dumping long prompts into SKILL.md. It encodes an iron law: every line must earn its tokens by improving consistency, reliability, or outcomes—otherwise delete it. The prescribed package shape keeps SKILL.md under roughly five hundred lines as the required workflow core, with optional scripts for deterministic work, references loaded only on demand, and assets that never bloat context. A copy-paste checklist guides creation, updates, refactors, debugging skill behavior, and packaging for distribution. Triggers cover plain-language requests like build a skill, improve a skill, or write SKILL.md, so solo builders can standardize how they capture domain procedures. Use it when you are investing in reusable agent capabilities across Validate scoping notes, Build features, and Operate iteration—not when you only need a single throwaway answer.900installs2SigmaSigma is a journey-wide agent skill that implements Bloom’s 2-Sigma one-on-one tutoring for solo builders learning through Claude Code, Cursor, Codex, Windsurf, Trae, and similar terminals. It adapts pacing with Socratic prompts, blocks progression until rubric scores show at least roughly eighty percent mastery, logs misconceptions, and brings prior topics back on a spaced schedule with interleaved questions. Visual artifacts help you see relationships between concepts without replacing your own reasoning. Use it whenever you must climb a learning curve—new framework before Build, compliance before Ship, distribution psychology before Launch—not only during initial research. Install via the documented skills.sh path under sanyuan0704/sanyuan-skills. It complements coding and review skills by building durable understanding rather than one-off fixes.881installs3Book StudyBook-study is an agent skill that supplies markdown templates for building a book-centric wiki: concept pages with definitions and applications, case pages with evidence and credibility notes, model pages with structure and when-not-to-use limits, and chapter summaries that tie arguments to linked knowledge points. It fits solo and indie builders who read business, technical, or strategy books and want durable, navigable notes instead of scattered highlights. Use it whenever you are distilling a book into a structured PKM layer your coding agent can append to, link across titles, and query during research, validation, or planning. The skill does not scrape books or run tools; it defines page shapes and editorial sections so outputs stay consistent and citable across your repo.1installs4Skill ReviewSkill-review is an agent skill reference that codifies how to judge other skills for Prism- and skills.sh-style quality. It loads during structured review Step 2 and spells out structure compliance—from mandatory name and description frontmatter through line-count limits, purposeful directories, and how scripts, references, and assets must tie back to SKILL.md with load instructions. The description-quality section teaches what good discovery copy looks like: multiple verb and noun variations, natural question-shaped triggers, and user-intent framing instead of implementation trivia, while flagging thin or keyword-starved descriptions. Red-flag guidance pushes oversized SKILL.md bodies into references, flattens over-nested docs, and calls out orphaned files or untested scripts. Solo builders publishing reusable agent capabilities use it to keep packages maintainable, citable, and install-worthy. It complements generic code review by focusing on procedural knowledge packaging rather than application security or PR diffs, and pairs naturally with brainstorming or writing-plans skills when you are iterating on a skill spec before merge.1installs5Wiki Ingestwiki-ingest gives solo builders and agent-assisted doc maintainers a rigid but fast path from raw material to interlinked wiki pages. Instead of free-form summaries, it prescribes markdown skeletons for concepts, products, architectural patterns, and A-versus-B comparisons—each with key points, context, related links, and source attribution so pages stay citable and browsable. That structure helps when code-review or research streams produce many half-formed notes and you need a single ingest shape the agent can repeat. Primary use is while building or refreshing documentation alongside a codebase, but the same templates support early idea research when you are cataloging competitors and patterns before commit. It does not fetch remote wikis or run ETL; it is template and section discipline for human or agent-authored ingest. Pair it with deeper review skills in the same repo when pages must reflect audited findings rather than generic placeholders.1installs