Skip to content

npow/claude-skills

Repository files navigation

claude-skills

License: Apache-2.0 Docs

Make Claude Code do complex multi-step workflows reliably with a single slash command.

How it works

Claude Code is powerful out of the box, but complex tasks — scaffolding a Python library, initializing a GitHub repo with CI/CD, diagnosing flaky tests, shipping a product idea end-to-end — require careful prompting and domain knowledge that gets lost between sessions. You end up re-explaining the same workflows, catching the same mistakes, and missing the same steps.

Skills fix that. Each skill is a directory of markdown files that Claude loads as context when you invoke it. The main SKILL.md defines workflow steps, golden rules, and a self-review checklist. Additional reference files cover specific phases — templates, patterns, examples — loaded only when needed. Because they live in ~/.claude/skills/, they're always available.

Skills

Building & Shipping

  • /build-python-library — Scaffolds pip-installable Python packages with src layout, pyproject.toml, and pytest, building core abstractions with adapters and verifying each module before proceeding.
  • /develop-board-game — Creates faithful digital board games as single-file HTML/JS/CSS apps, handling rules, multiplayer, and Playwright tests.
  • /init-github-repo — Initializes repos with a JTBD-focused README, CI/CD workflow, license, .gitignore, and SEO-optimized description and topics.
  • /ship-it — Takes a product idea through design, build, test, integrate, and package phases using parallel subagents with quality gates between each phase.

Research & Design

  • /competitive-matrix — Researches a market and renders an interactive, color-coded comparison matrix in the browser.
  • /deep-design — Adversarially stress-tests a design with parallel critic agents across dimensions like security, scalability, and UX. Iterates until flaws are saturated. Output is a battle-tested spec.
  • /deep-idea — Generates genuinely novel ideas via five orthogonal forcing functions and adversarial novelty killing. Forces extrapolation over interpolation.
  • /deep-qa — Audits any artifact (document, spec, code, skill) for defects using parallel critic agents across QA dimensions tailored to the artifact type.
  • /deep-research — Systematic multi-dimensional research using parallel agents across orthogonal dimensions (WHO/WHAT/HOW/WHERE/WHEN/WHY). Source quality tiers, spot-checked verification, honest coverage report.
  • /gap-finder — Finds viable product or business ideas through adversarial generation-and-kill cycles, validating each candidate against real market data.
  • /proposal-reviewer — Critically reviews proposals by fact-checking claims against primary sources, mapping the competitive landscape, and identifying structural flaws.
  • /research-brief — Produces a structured, cited research brief with counterevidence and synthesis from web sources.
  • /spec — Turns a conversation or rough idea into a complete, structured technical specification saved as a markdown file.

Visualization

  • /chart — Renders interactive Chart.js charts (bar, line, pie, radar, time series) live in the browser from context-supplied data.
  • /diagram — Renders interactive, draggable node-and-edge diagrams in the browser using Cytoscape.js.
  • /slides — Converts content into a Reveal.js slide deck and opens it live in the browser.
  • /table — Renders sortable, filterable comparison tables live in the browser with zero dependencies.
  • /timeline — Renders interactive Gantt-style timelines and roadmaps in the browser using vis-timeline.

Debugging & Quality

  • /flaky-test-diagnoser — Runs multi-run experiments, isolation tests, ordering permutations, and timing analysis to find the root cause of a flaky test.
  • /prod-readiness — Scans codebases for 24 production readiness items and produces a scored report with pass/fail/warning and suggested fixes.

Meta

  • /create-skill — Builds new Claude Code skills following harness engineering best practices: table-of-contents architecture, golden rules, self-review loops, progressive disclosure.
  • /magic-fetch — Logs capability gaps in real-time. When Claude hits a wall, it records exactly what tool, API, or access would have solved it — building a prioritized integration roadmap over time.

Install

# Clone the repo
git clone https://github.com/npow/claude-skills.git

# Option A: Symlink individual skills
ln -s "$(pwd)/claude-skills/build-python-library" ~/.claude/skills/build-python-library

# Option B: Use the repo as your skills directory
# (back up any existing skills first)
mv ~/.claude/skills ~/.claude/skills.bak
ln -s "$(pwd)/claude-skills" ~/.claude/skills

Usage

Invoke any skill by typing its name as a slash command in Claude Code:

/deep-research
/ship-it
/flaky-test-diagnoser

Contributing

Skills live directly in this repository. To contribute:

  1. Fork the repository
  2. Use /create-skill to scaffold and test your new skill
  3. Submit a PR

See create-skill/SKILL.md for the complete guide.

License

Apache 2.0

About

Reusable slash-command workflows that make Claude Code reliable on complex multi-step tasks.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors