ClaudeSkills Agent Skill

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

<p align="center"> <img src="assets/social/agent-skills-concept-hero.png" alt="Agent Skills concept: experience becomes reusable capability" width="100%"> </p>

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:

<details> <summary><b>Show them</b></summary>

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

</details>

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:

<p align="center"> <img src="skills/Geek-skills-deck-studio/style-library/creative/bauhaus-preview.png" alt="Bauhaus deck style preview" width="24%"> <img src="skills/Geek-skills-deck-studio/style-library/creative/constructivist-preview.png" alt="Constructivist deck style preview" width="24%"> <img src="skills/Geek-skills-deck-studio/style-library/media/neubrutalism-preview.png" alt="Neubrutalism deck style preview" width="24%"> <img src="skills/Geek-skills-deck-studio/style-library/business/aicher-preview.png" alt="Aicher deck style preview" width="24%"> </p>

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.

<details> <summary><b>What each check proves, and what it doesn't</b></summary>

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.

</details>

Install options

<details> <summary><b>Show them</b></summary>

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.

</details>

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.

License

MIT