Skip to content

Sanyuan Skills - Setup Guide

Source: sanyuan0704/sanyuan-skills via skills.sh - 30.0K combined installs across 6 indexed listings; first seen Oct 9, 2026 (extended-query probe) GitHub: sanyuan0704/sanyuan-skills (3,939 stars, MIT license; pushed May 11, 2026; skills/<name>/SKILL.md layout) Category: Agent Workflow / Code Review Quality Tier: 🟡 Beta - 3,939-star MIT collection (Sanyuan); production-grade skill set; all sampled verdicts Pass

Sanyuan (sanyuan0704), known for in-depth writing on React internals, publishes this collection of production-grade agent skills for Claude Code and other agent terminals. Six skills cover two loops a working engineer repeats constantly. The first is code review: code-review-expert at 13,310 installs runs a senior-engineer pass over a diff covering SOLID, security, performance, error handling, and boundary conditions. The second is learning and library upkeep: sigma and book-study run Socratic tutoring with mastery checks, wiki-ingest compiles sources into a cross-referenced knowledge base, and skill-forge plus skill-review create and audit skills.

The collection is deliberately small and composable. Every skill lives at skills/<name>/SKILL.md and installs independently, so you can adopt the code-review pass without taking the tutoring stack. All six listings are above 2,100 installs, and the README ships a per-skill install command for each one.

Installation

Prerequisites: Node.js, since every skill installs through the npx skills CLI.

# Install the full collection
npx skills add sanyuan0704/sanyuan-skills

The README ships a per-skill install command for each skill, useful when you only need one:

npx skills add sanyuan0704/sanyuan-skills --path skills/code-review-expert
npx skills add sanyuan0704/sanyuan-skills --path skills/sigma
npx skills add sanyuan0704/sanyuan-skills --path skills/skill-forge
npx skills add sanyuan0704/sanyuan-skills --path skills/book-study
npx skills add sanyuan0704/sanyuan-skills --path skills/wiki-ingest
npx skills add sanyuan0704/sanyuan-skills --path skills/skill-review

Once installed, invoke each skill from your agent terminal:

/code-review-expert    # Review current git changes
/sigma <topic>         # Start a tutoring session
/skill-review          # Audit an existing skill's quality
/skill-forge           # Create a new skill
/wiki-ingest           # Compile content into a wiki knowledge base
/book-study <book>     # Start a guided reading session

What It Provides

Install counts from the Oct 10, 2026 skills.sh snapshot:

Skill Installs Use For
code-review-expert 13,310 Senior-engineer code review covering SOLID, security, performance, error handling, and boundary conditions
sigma 4,661 1-on-1 AI tutor built on Bloom's 2-Sigma mastery learning with Socratic questioning
skill-forge 4,000 Meta-skill for creating high-quality skills with 12 battle-tested techniques
book-study 3,276 Reading coach with knowledge compilation, Socratic mastery testing, spaced repetition, and cross-book querying
wiki-ingest 2,587 Compile articles, documents, or notes into a structured, cross-referenced wiki knowledge base
skill-review 2,135 Quality audit for skills: structure, description, workflow, token efficiency, and anti-patterns

All 6 indexed listings clear the 500-install threshold and appear in the table above.

Why This Matters for Hermes Agents

Code review is one of the highest-leverage places to put an agent, and code-review-expert encodes the checklist a senior engineer applies on every diff instead of offering generic commentary. The skill-library half matters just as much for teams now maintaining dozens of their own skills: skill-forge packages 12 techniques for writing new ones well, and skill-review audits existing skills for structure, token efficiency, and anti-patterns before they get shared. The learning skills (sigma, book-study, wiki-ingest) turn the same agent into a patient tutor and a knowledge compiler, which fits long research sprints where context has to be rebuilt fast. Everything installs per-skill through the standard npx skills add flow, so a Hermes agent can pull in just code-review-expert and grow from there. The repo is small, MIT, and readable end to end, which makes it a useful study model for teams writing their own production skills.

Usage

You say What happens
"Review my staged changes before I commit" code-review-expert runs a senior review pass over the diff: SOLID, security, performance, error handling, boundary conditions
"Teach me React Server Components from scratch" sigma starts a Socratic tutoring session and tests mastery as you go
"Audit the skill we just wrote" skill-review checks structure, description, workflow, token efficiency, and anti-patterns
"Draft a skill for our deploy runbook" skill-forge applies its 12 techniques to produce a well-structured new skill
"Turn these API docs and notes into a wiki" wiki-ingest compiles the mixed sources into a cross-referenced knowledge base
"Coach me through this book in two weeks" book-study schedules the reading, tests mastery, and links ideas across books

Verification

# Confirm the install through the skills CLI
npx skills list | grep -i sanyuan

# Or check the skills directory your agent scans
ls ~/.claude/skills/ | grep -E 'code-review-expert|skill-forge'

# Review the flagship skill straight from GitHub before installing
curl -sL https://raw.githubusercontent.com/sanyuan0704/sanyuan-skills/main/skills/code-review-expert/SKILL.md | head -20

Security

skills.sh verdicts for sampled skills (verified Oct 10, 2026) - re-check the security pages on skills.sh before production use:

Skill Gen Agent Trust Hub Socket Snyk
code-review-expert Pass Pass Pass
code-review - - -
codex-review - - -

The scored sample passes on all three engines, but coverage is partial: only code-review-expert has verdict rows; the tutoring and skill-authoring skills were not sampled.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations

  • The collection comes from a single independent author; it is six focused skills rather than a broad framework.
  • Maintenance cadence: the repo was last pushed May 11, 2026, so it predates the newest agent-platform conventions.
  • Verdict coverage is partial; extend the security review to the unsampled skills before production use.
  • No MCP servers, hooks, or automation layer ship with the collection; these are plain SKILL.md skills.
  • Snapshot data, verified Oct 10, 2026: 29,969 combined installs across 6 indexed listings; 3,939 GitHub stars; MIT; last pushed May 11, 2026. Counts drift over time.

Pro Tips

  • Start with code-review-expert alone; run it on staged diffs before commits and calibrate expectations from its feedback.
  • Pair the authoring loop: draft with skill-forge, then press skill-review on the result before sharing it.
  • For sigma, state your current level and target depth in the prompt; the mastery checks adapt better with that context.
  • wiki-ingest takes mixed sources (articles, docs, raw notes); feed it captures and let it build the cross-references.
  • Install per-skill with --path when you only need one; the collection is small, but there is no reason to carry skills the agent will not use.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing