agent skills for agentic coding tools. extremely opinionated. updated (almost) daily.
| plan | executable implementation plans |
| code | implement plans with quality gates |
| orch | multi-agent parallel execution |
| pr-review | code review — correctness, security, quality |
| qual | multi-lens analysis with specialist agents |
| refactor | architectural refactoring with parallel agents |
| distill | maximize LOC reduction |
| ux | UI evaluation across Nielsen's heuristics |
| qg | quality gates — format, lint, typecheck, tests, build |
| commit | conventional commits + push |
| pr | PR from worktree branch |
| worktree | git worktree for isolated dev |
| rebase | rebase branch onto target, resolve conflicts |
| worktree-clean | remove completed worktree + branch |
| timesheet | monthly work summary from git |
| find-skills | discover and install from registries |
| improve-skill | assess skills against ecosystem + competing tools |
| skill-creator | create, modify, benchmark, eval skills |
| intern | analyze sessions, detect patterns, auto-improve |
| prompt | design production-grade LLM prompts |
| transformer | rewrite skills for reasoning model performance |
| design | divergent exploration with independent reasoning agents |
| seo-geo | SEO + generative engine optimization |
| pen-design | design UIs in .pen files — design thinking + visual systems |
| pen-code | convert between .pen designs and React/Tailwind code |
| approve-piped-bash | auto-approves piped commands when every segment is already in your allowlist |
| detect-skill-invocation | type /skillname in a prompt, get the skill injected into context |
Skills work out of the box, but some features require settings.json entries:
| Setting | Why |
|---|---|
env.CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS: "1" | Required for skills that spawn agent teams (orch, qual, refactor, intern, design) |
permissions.additionalDirectories: ["~/.claude/skills"] | Lets teammates read skill files — needed if you symlink skills into ~/.claude/skills/ |
Hook scripts ship in hooks/ but must be wired in settings.json under the hooks key to take effect — see Claude Code docs on hooks.
# clone and symlink (recommended — stays updated)
git clone git@github.com:JHostalek/dotclaude.git ~/dotclaude
ln -s ~/dotclaude/skills/* ~/.claude/skills/
# or just copy what you need
cp -r skills/pr-review ~/.claude/skills/Some skills were created by other authors and are included under their original licenses — not CC0.
| Skill | Author | License | Source |
|---|---|---|---|
| skill-creator | Anthropic | Apache 2.0 | anthropics/skills |
| seo-geo | ReScienceLab | MIT | ReScienceLab/opc-skills |
Follows the Agent Skills open standard — works with Claude Code, Codex CLI, Cursor, GitHub Copilot, and anything that reads SKILL.md.
CC0 1.0 — public domain. Excludes third-party skills listed above.
dotclaude is an open-source, actively maintained collection of production-ready Agent Skills for Claude Code, Anthropic's agentic AI coding CLI. Skills are reusable prompt-based modules that extend what Claude Code can do — each packaged as a SKILL.md file with YAML frontmatter following the Agent Skills open standard adopted by Anthropic, Microsoft, OpenAI (Codex CLI), Cursor, GitHub Copilot, Atlassian, and Figma.
Claude Code ships with general-purpose intelligence but no opinionated workflows. dotclaude adds battle-tested skills for the full development lifecycle: planning (plan), implementation (code), code review (pr-review, qual), refactoring (refactor, distill), git workflow (commit, pr, rebase, worktree, worktree-clean), team orchestration (orch), and SEO/GEO optimization (seo-geo). Each skill encodes specific methodologies — the pr-review skill checks for correctness, security, and quality; the design skill spawns independent agents with different reasoning methods to prevent convergence bias; the intern skill analyzes past sessions to detect recurring patterns and auto-improve your configuration; the seo-geo skill applies Princeton GEO research methods and DataForSEO API integration to optimize for both traditional and AI search engines.
| Feature | dotclaude | anthropics/skills | awesome-claude-skills |
|---|---|---|---|
| Actively maintained | Yes (updated (almost) daily) | Occasional | No (curated links) |
| Agent team orchestration | Yes (orch, qual, refactor, intern) | No | No |
| Multi-agent design exploration | Yes (design skill) | No | No |
| Skill meta-tooling | Yes (skill-creator, improve-skill, find-skills) | No | No |
| SEO/GEO optimization | Yes (seo-geo: free audit + 9 DataForSEO scripts) | No | No |
| Git workflow automation | Yes (commit, pr, rebase, worktree, worktree-clean) | No | Partial |
| Design-to-code (Pencil) | Yes (pen-design, pen-code) | No | No |
| Self-improving configuration | Yes (intern skill) | No | No |
| Agent Skills standard compliant | Yes | Yes | Varies |
| Cross-tool compatible | Yes (Codex, Cursor, Copilot) | Claude Code only | Claude Code only |
Build skills start the development loop: plan creates decomposed implementation plans with verification criteria; code executes plans with incremental testing and legacy cleanup; orch orchestrates multi-agent teams for parallel implementation.
Quality skills catch issues before they reach production: pr-review performs multi-dimensional code review (correctness, security, quality); qual spawns specialist agent teammates that analyze code from different angles then consolidate findings; refactor runs comprehensive architectural analysis with parallel agent teams per module; distill maximizes line-of-code reduction while maintaining readability; ux evaluates user interfaces across 13 dimensions mapped to Nielsen's heuristics; qg runs all quality gates (format, lint, typecheck, tests, build) and reports PASS/FAIL per gate.
Ship skills handle the git workflow from branch to PR to merge: commit enforces conventional commits with branch guards; pr creates pull requests with conventional formatting; worktree and worktree-clean manage git worktree lifecycle for isolated feature development; rebase rebases branches onto targets and resolves conflicts; timesheet generates monthly work summaries from git history.
Improve skills level up your Claude Code setup itself: find-skills discovers installable skills from community registries; improve-skill compares your skills against the open ecosystem and competing AI coding tools; skill-creator creates, modifies, benchmarks, and evaluates skills with variance analysis; intern analyzes recent Claude Code sessions to detect recurring patterns and automatically write skill and rule improvements; prompt designs production-grade LLM prompts through interactive collaboration; transformer rewrites skills so reasoning models perform them better — strips scaffolding, preserves domain knowledge, adds calibration.
Domain skills solve specific problems: design runs divergent design exploration by spawning independent agents with clean contexts and different reasoning methods (morphological analysis, TRIZ, constraint relaxation, biomimicry) to prevent convergence bias; seo-geo handles SEO and Generative Engine Optimization for both traditional search engines (Google, Bing) and AI search engines (ChatGPT, Perplexity, Gemini, Google AI Overview, Microsoft Copilot, Claude) — includes a free technical SEO audit script, 9 DataForSEO-powered Python scripts (keyword research, related keywords, autocomplete ideas, SERP analysis, backlinks, domain overview, competitor gap analysis), 6 reference guides (Princeton GEO research, platform-specific ranking algorithms, JSON-LD schema templates, SEO audit checklist, tools and APIs), and a real-world optimization case study.
Experimental — Pencil integration: pen-design creates production-quality UI designs in Pencil .pen files with design thinking and visual systems methodology; pen-code converts between .pen designs and React/Tailwind code (export from design or sync changes back). These skills require the Pencil desktop app and its MCP server — see AI integration docs for setup.
Install by copying skill directories into ~/.claude/skills/ or by cloning the repository and symlinking. Each skill is a self-contained directory with a SKILL.md file (YAML frontmatter defining name, description, and tool restrictions) plus optional supporting files in references/, scripts/, examples/, and agents/ subdirectories. Skills are invoked via /skillname slash commands in Claude Code.
What is dotclaude?
dotclaude is an actively maintained collection of Agent Skills for Claude Code that add extremely opinionated workflows for planning, implementation, code review, refactoring, git operations, team orchestration, SEO/GEO optimization, and self-improving configuration. Each skill follows the Agent Skills open standard and works across Claude Code, Codex CLI, Cursor, and VS Code Copilot.
How do I install dotclaude skills?
Clone the repository and copy or symlink individual skill directories into ~/.claude/skills/. Each skill is self-contained — no dependencies, no build step, no configuration beyond placing the directory. Skills are automatically discovered by Claude Code on next session start.
Do these skills work with tools other than Claude Code?
Yes. All skills follow the Agent Skills open standard (agentskills.io) adopted by Anthropic, Microsoft (VS Code Copilot), OpenAI (Codex CLI), Cursor, GitHub, Atlassian, and Figma. Any tool that reads SKILL.md files with YAML frontmatter can use these skills.
What makes dotclaude different from other skill collections?
Three things: (1) agent team orchestration — skills like orch, qual, refactor, and intern spawn multiple agents working in parallel; (2) self-improving meta-tooling — skill-creator, improve-skill, and intern let you create, benchmark, and auto-improve skills; (3) depth — each skill encodes specific methodologies rather than generic checklists (e.g., seo-geo includes 9 DataForSEO scripts and 6 reference guides based on Princeton GEO research).
What does the seo-geo skill do?
The seo-geo skill optimizes websites for both traditional search engines (Google, Bing) and AI search engines (ChatGPT, Perplexity, Gemini, Google AI Overview, Microsoft Copilot, Claude). It includes a free technical SEO audit script (no API key needed), 9 Python scripts powered by the DataForSEO API for keyword research, SERP analysis, backlink analysis, domain overview, competitor gap analysis, related keywords, and autocomplete ideas. Reference guides cover the 9 Princeton GEO optimization methods, platform-specific ranking algorithms for each AI search engine, JSON-LD schema templates, and a comprehensive SEO audit checklist.