skills

Claude Code Skills

Ready-to-use Claude Code skill files. Each skill is a drop-in directory you copy to your .claude/skills/ folder.

How to Use

Install a single skill

Copy the skill directory to your project or personal skills folder:

# Project-scoped (only this project)
cp -r skills/commit .claude/skills/

# Personal (available in all your projects)
cp -r skills/commit ~/.claude/skills/

Install all skills at once

# Project-scoped
cp -r skills/. .claude/skills/

# Personal
cp -r skills/. ~/.claude/skills/

After copying, the skill is immediately available. Type /skill-name to invoke it directly, or let Claude pick it up automatically when relevant.

Skill scopes

Scope	Path	Who can use it
Personal	~/.claude/skills/<name>/SKILL.md	You, across all projects
Project	.claude/skills/<name>/SKILL.md	Everyone working in this project

---

Skills Reference

Skill	Invoke	Type	Description
commit	/commit	Task	Run tests, generate a conventional commit message, and commit staged changes
pr	/pr [base]	Task	Push branch and create a pull request with generated title and description via gh
review	/review	Task	Review the current diff for bugs, security issues, and style problems
fix-issue	/fix-issue <number>	Task	Read a GitHub issue, implement the fix, write tests, and commit
tdd	/tdd <feature>	Task	Red-green-refactor TDD cycle: failing test first, then implement, then clean up
refactor	/refactor <target>	Task	Refactor a file or function while preserving behavior, with test verification
explain	/explain [target]	Research	Explain code with an analogy, ASCII diagram, and step-by-step walkthrough
security-scan	/security-scan [path]	Research	Audit the codebase for OWASP Top 10 vulnerabilities with a structured report
api-conventions	(auto-loaded)	Reference	REST API design conventions applied whenever writing or reviewing API code
doc-gen	/doc-gen <target>	Task	Generate or update inline documentation and README files for a module
migrate	/migrate <path> <target>	Task	Migrate code from one framework or pattern to another, preserving behavior
debug	/debug <error>	Task	Systematic debugging: reproduce, isolate with hypothesis testing, fix, verify
onboard	/onboard	Research	Generate a codebase overview with architecture diagram, key files, and conventions
changelog	/changelog [version]	Task	Generate a categorized changelog from git history since the last release tag
perf-audit	/perf-audit [target]	Research	Find performance bottlenecks: N+1 queries, slow algorithms, memory issues

Types explained

• Task: has side effects (writes files, runs commands, creates commits). Most require explicit invocation via /skill-name. Claude will not trigger them automatically.
• Research: runs in a forked subagent context with read-only or limited tools. Returns a summary to your main conversation.
• Reference: background knowledge loaded into Claude's context automatically when relevant. Not a command you invoke - Claude reads it silently.

---

Skill Details

commit

Stages all changes, runs the project test suite, and generates a Conventional Commits-formatted message (type(scope): summary). Aborts if tests fail. Requires disable-model-invocation: true so Claude cannot auto-commit your work.

pr

Detects the base branch, pushes the current branch, generates a PR title and description (with Summary, Changes, Testing, and Related Issues sections), and runs gh pr create. Falls back to a manual URL if gh is not installed.

review

Runs in a forked Explore agent. Fetches the current diff and analyzes it for correctness, security, error handling, performance, and style. Produces a structured report with severity-rated findings and a final verdict.

fix-issue

Reads a GitHub issue with gh issue view, locates the relevant code, writes a failing test first, implements the fix, verifies tests pass, and commits with Fixes #N in the message.

tdd

Guides a complete red-green-refactor cycle. Enforces the discipline of writing the failing test before any implementation. Works with any test framework the project already uses.

refactor

Requires passing tests as a precondition. Makes only behavior-preserving changes (renaming, extracting helpers, flattening nesting, replacing magic values). Runs tests after each step to catch regressions early.

explain

Runs in a forked Explore agent. Always produces three things: an everyday-life analogy, an ASCII architecture or flow diagram, and a "gotcha" - the common misconception about the code. Auto-invoked when you ask "how does this work?"

security-scan

Systematically checks for all OWASP Top 10 categories using Grep and Read. Reads surrounding code to rule out false positives before reporting. Output includes severity, file location, evidence, and remediation for each finding.

api-conventions

A reference skill (user-invocable: false) that Claude loads silently when you write or review API code. Covers kebab-case URLs, camelCase JSON, HTTP status codes, cursor pagination, versioning, authentication headers, idempotency keys, and rate limiting headers.

doc-gen

Detects the project's documentation format (JSDoc, Google docstrings, Go doc comments) and writes inline documentation plus module-level READMEs. Uses test files to infer intended behavior when source comments are absent.

migrate

Builds a source-to-target concept mapping before touching any files, then migrates file-by-file with test verification after each file. Flags constructs with no direct equivalent rather than making silent assumptions.

debug

Enforces a three-phase process: reproduce (confirm the bug is consistent), isolate (form and test at least three hypotheses), fix (minimum change, full test suite). Prevents the common trap of fixing symptoms without finding root causes.

onboard

Runs in a forked Explore agent. Reads README, CLAUDE.md, entry points, and CI config to produce a guide covering: what the project does, an ASCII architecture diagram, a directory map table, an ordered "read these first" file list, local setup steps, conventions, and a "where to make common changes" section.

changelog

Parses conventional commit prefixes to group changes into Features, Bug Fixes, Performance, Docs, and Maintenance sections. Suggests a semver bump (major/minor/patch) based on the commit types found. Prepends the entry to CHANGELOG.md.

perf-audit

Runs in a forked Explore agent. Covers six categories: algorithmic complexity, database/I/O patterns, memory allocation, concurrency, frontend rendering, and caching opportunities. Separates findings into "Quick Wins" (low-risk) and "Larger Investments" (requires design changes).