ClaudeSkills
ClaudeSkills is a collection of Agent Skills: 13 skill packages written in the open SKILL.md format. Each skill packages the complete process for getting something done — steps, templates, scripts, examples, and acceptance checks are all files — and installs into any client that supports Agent Skills.
A prompt lets a model do something once, then dies with the conversation; a skill keeps the method around, so it can be reused, inspected, and improved.
Contents: Quick start · Skills · Verification · Install options · Contributing
Website · Release 1.0.1 notes · Security
Quick start
Install one and try it — deck-studio, for example:
git clone --depth 1 https://github.com/staruhub/ClaudeSkills.git && cd ClaudeSkills
python3 scripts/install_skill.py deck-studio
Then just tell your agent:
Use deck-studio to turn this quarterly review into an 8-slide consulting deck
Skills install to ~/.agents/skills/ by default, which most clients scan. For other ways to install, see Install options.
Skills
Four flagships:
| Skill | What it does |
|---|---|
deep-research |
Research: scopes the question, gathers sources, registers each one, checks citations; hands back a report with references and limits |
product-manager |
grill-me-to-doc: reads the repo first, asks one question per round, turns a vague idea into a product document. No code before the doc is approved; interruptions resume |
deck-studio |
Decks: outline, per-page briefs, registered layouts, visual checks; delivers content or per-slide visuals |
wechat-article-writer |
WeChat articles: copy, image prompts, layout HTML — solo or as a pipeline. Never publishes for you |
Nine professional skills:
| Skill | What it does |
|---|---|
pair-programming |
Writes code with structured self-review aimed at common AI-code defects |
security-audit |
Reviews code and dependencies for security issues |
solution-architect |
System design, tech selection, architecture review |
threejs-performance |
Diagnoses and tunes Three.js performance |
mineru-pdf-parser |
Turns PDFs into Markdown or JSON with a local MinerU install |
ai-sales-champion |
Explains technical capability as business value a customer understands |
keqian-method |
A single-agent, SDD, quality-gated product development method |
xuefeng-method |
An open-behavior, model-driven AI-native product method |
c-drive-cleaner |
Windows C-drive cleanup, dry-run by default |
Experimental ones (exam prep, weather reports, podcast generation, and friends) live in lab/. They don't count toward the curated set and skip the same checks.
Verification
Everything checkable is a script — run them yourself:
python3 scripts/validate.py # structure checks for the 13 curated skills
python3 scripts/run_routing_evals.py # 91 routing cases
python3 tests/task_b/run_contract_tests.py # 17 contract cases across the four flagships
You can also inspect the artifacts directly. deck-studio keeps its generators, rendered pages, rubrics, and review notes in the repo. Four of the 17 styles:
Examples: Constructivist · Moshiro consulting report · Yinghuang bootcamp proposal · Polar Night AI Native
Self-test scores: the Constructivist example scored 7.1/10 on the repo's own rubric; a position-swapped three-judge comparison came out 42.3 to 29.7. The process and data are in the repo — rerun them yourself.
| Check | Proves | Doesn't prove |
|---|---|---|
run_contract_tests.py |
17 fixed cases covering the success, resume, and failure paths of all four flagships; Deck additionally runs real Chrome rendering and PPTX assembly | Output quality on a different model, real image generation, actual WeChat publishing |
validate.py |
Directory structure of the 13 curated skills matches the repo contract | Real-business E2E for every skill |
run_routing_evals.py |
Schema, target, uniqueness, and conflict checks pass for 91 routing cases across 10 skills | Routing accuracy when a real model runs it |
| Python / Node compile checks | The repo's 13 Python files and 7 JavaScript files all parse | Network, external tools, production availability |
Full record in verification/2026-07-31/README.md.
Install options
python3 scripts/install_skill.py --list # list all short names
python3 scripts/install_skill.py deep-research # install any skill
python3 scripts/install_skill.py deep-research --project # current project only
python3 scripts/install_skill.py deep-research --client claude-code # Claude Code's directory
cp -r skills/Geek-skills-deep-research ~/.agents/skills/deep-research # manual copy; the directory name is the skill name
git pull && python3 scripts/install_skill.py deck-studio --force # update
rm -rf ~/.agents/skills/deck-studio # uninstall
FAQ:
- Installed but not found: check whether your client scans
.agents/skills/; if not, use its native directory. - No automatic trigger: name the skill in your request. Triggers vary by client — some match natural language, others want slash commands.
- Re-install after
git pull— installed skills are copies.
Contributing
Found a bug or built something with a skill? Open an issue with redacted input, output, and repro steps. To add a skill, read CONTRIBUTING.md: new work starts in lab/ and graduates after passing the checks.
If it saved you time, leave a star. The WeChat promo image and ready-to-send copy are in assets/social/ — take them.
Security
What each skill reads, whether it touches the network, runs commands, or handles credentials — per skill in SECURITY.md.