Open source · MIT
Agent skills
Skills I wrote for my own work with coding agents. Each one is a folder with a SKILL.md, so it runs in Claude Code, OpenCode, Codex and any other agent that reads the format. Pick one and copy the install command.
- productivityv1.1.1
mac-cleanup
Analyze and safely free up disk space on macOS: measures what uses storage, cleans only caches that rebuild themselves, finds leftovers of uninstalled apps and guides the removals that need your judgment.
macOS 13 or later.3 scripts3 references - productivityv1.0.0
race-photo-edit
Cull and edit running and street-race photos for sale (Foco Radical, Fotop): a script measures exposure, asphalt color cast, sharpness, EXIF and bursts and builds contact sheets; the agent rejects what will not sell, picks the best of each burst, groups by light, assigns presets with per-photo corrections, crops and optional background blur, writes every approved photo to an upload-ready folder without touching originals, and gives shooting feedback.
1 script3 references - developmentv4.2.0
crew
Run a feature, bug fix or refactor through a fast crew: the orchestrator writes the spec, contracts and a task board, fans implementation and tests out to up to 10 parallel Haiku workers (a bundled crew-worker agent), integrates, runs the checks and reviews, with Haiku QA and an independent Reviewer for large or risky changes, sized runs (small work goes solo) and capped fix loops.
1 script3 references - productivityv1.1.0
handoff
End a session and continue in a fresh one without losing context: writes a self-contained handoff (goal, verified state, next steps, decisions, dead ends, user preferences, files to read, running processes, git snapshot) to .claude/handoffs/ or .agents/handoffs/ (in the project for coding projects, in the user's home otherwise), copies a short start prompt to the clipboard, and resumes by checking the handoff against the current code and git state.
3 scripts1 reference - developmentv1.0.0
docs-sync
Create and update README.md and AGENTS.md so they match the code: a stdlib script inventories the repo and finds drift (dead commands, broken links and paths, stale env vars, oversized or duplicated agent files, what changed since each doc), then the skill makes the smallest accurate edit, keeps one home per fact (README for humans, AGENTS.md for agents, CLAUDE.md importing it), prunes instead of appending, never invents commands and verifies before reporting.
1 script3 references - marketingv1.1.0
seo-geo
Improve SEO and visibility in AI chats (ChatGPT, Perplexity, Gemini, AI Overviews, Claude, Copilot): crawler-eye audit (robots and CDN blocks per AI bot, JS-only rendering, indexability, JSON-LD), a full-site crawler (broken links, redirects, duplicates, orphans, depth, sitemap gaps), backlink analysis from Search Console/Bing/Ahrefs/Semrush exports with links to reclaim and a competitor link gap, a prompt set measuring citations and wrong claims, Search Console quick wins, code fixes, quotable content and a prioritized report.
3 scripts4 references - developmentv1.0.0
pentest
White-box security audit for your own web app or API: read-only scan plus OWASP Top 10 review of code, auth, deps and headers, with a ranked report and fixes.
1 script2 references - productivityv1.0.0
product-naming
Find a brandable product or company name and check it can be owned: generates candidates in batches by strategy, bulk-checks domains straight from the registries (self-tested RDAP/whois), filters by meaning and sound in every market language, scores fit and marketing potential, and vets finalists (who owns a taken .com, search conflicts, USPTO/INPI/WIPO pre-screen, social handles).
3 scripts3 references
Every skill here comes straight from the repository. When I push a new one, it shows up on this page.