Thirteen curated Agent Skills package the steps, templates, scripts, examples, and quality gates behind repeatable work. They follow the open SKILL.
Overview
Thirteen curated Agent Skills package the steps, templates, scripts, examples, and quality gates behind repeatable work. They follow the open SKILL.
README
Four flagship workflows, rebuilt
| You ask | The workflow does | You get |
|---|---|---|
| “Compare three RAG architectures for our support team.” | 🔬 deep-research scopes the question, gathers multiple sources, maintains a source registry, and checks citations |
A decision memo or full report with conclusions, evidence, trade-offs, and explicit limitations |
| “Don’t code yet. Ask me one question at a time until the product is clear.” | 📋 product-manager enters grill-me-to-doc: it reads repository evidence first, asks one decision per turn, and can resume after interruption |
A reviewable PRODUCT-DOC, decision log, and open questions—with a hard stop before implementation |
| “Turn this quarterly review into a consulting-style deck.” | 🎞️ deck-studio confirms the outline before page briefs, registered layouts, and visual QA |
Deck content, per-slide visuals, or an infographic set, depending on the delivery mode |
| “Turn these notes into a WeChat article, including image prompts and layout.” | ✍️ wechat-article-writer can run article, image-prompts, layout, or the full pipeline |
An article, provider-neutral image-prompt manifest, and WeChat-safe inline HTML—without pretending prompts are images or auto-publishing |
Each flagship exposes the work between request and deliverable: when it asks, which references it loads, what it validates, and where it stops. You are installing an inspectable workflow, not a magic sentence.
🚀 Install in 30 seconds
Start with one skill you will actually use:
git clone --depth 1 https://github.com/staruhub/ClaudeSkills.git && cd ClaudeSkills
python3 scripts/install_skill.py deck-studio
Then ask any skills-compatible agent:
Use deck-studio to turn this quarterly review into an 8-slide consulting deck
The installer defaults to the cross-client convention at ~/.agents/skills/deck-studio. Replace deck-studio with deep-research, product-manager, or wechat-article-writer to install another flagship.
Not another prompt dump
| A disposable prompt | ClaudeSkills |
|---|---|
| The conversation ends; next time you start over | Workflows, templates, and references live in version control |
| The model improvises every intermediate step | Inputs, stages, stop conditions, and deliverables are explicit |
| An interruption can erase decisions | Stateful workflows write resumable handoff files, including grill-me-to-doc and research delta updates |
| “Looks good” is the only test | Deterministic parts use schemas, scripts, fixtures, and negative cases; subjective quality stays subject to review |
| You discover file, network, or shell access after install | SECURITY.md discloses reads, writes, network, commands, credentials, and deletion per skill |
Actual capabilities still depend on the tools and permission model of the host agent. This repository makes the workflow and its boundaries inspectable; it does not bypass client permissions or market static checks as production performance.
Try one in 60 seconds
| If you are working on | Start with | Inspect first |
|---|---|---|
| Technical, competitive, or policy research | Use deep-research to compare… and produce a cited decision memo |
Research workflow and artifacts |
| Turning an early idea into a product document | Use product-manager in grill-me-to-doc mode: I want to build… |
One-question interview contract |
| A presentation, pitch, or training deck | Use deck-studio to turn… into a 10-slide …-style deck |
Delivery modes and layout system |
| WeChat copy, image prompts, and inline HTML | Use wechat-article-writer full-pipeline to turn… into… |
Four execution modes |
Do not install all 13 at once. Run one workflow against a real task this week; keep it if it earns the slot.
Inspect the artifacts before you install
Deck Studio keeps generators, rendered pages, rubrics, and judge feedback in the repository. Here are four of its 17 rendered style seeds:
Constructivist · Moshiro consulting report · Yinghuang bootcamp proposal · Polar Night AI Native
In the repository’s blind model-based self-evaluation, the Constructivist example scored 7.1/10 on the documented rubric; a position-swapped three-judge comparison scored the current pipeline 42.3 vs 29.7. These are reproducible project artifacts, not third-party certification.
📚 All Skills
Flagships (4)
deck-studio · deep-research · product-manager · wechat-article-writer
Lab
Personal and experimental workflows—exam prep, weather reports, image and podcast generation, and A-share analysis—live in lab/. They are not part of the 13-skill curated set or its gates.
Help make it better
Found a bug, or built something with a skill? Open an issue with a redacted input, output, and reproduction steps when possible. To contribute a skill, start with CONTRIBUTING.md: new work incubates in lab/ and graduates after meeting the repository gates.
If a workflow genuinely saved you time, leave a ⭐ and send it to someone still rebuilding the same prompts every week. Stars help discovery; they are not a quality certificate.
The WeChat launch poster and ready-to-send copy live in assets/social/.
License
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add staruhub/claudeskills