My preferred agent skills — a personal, curated, cross-agent collection. SKILL.md is an open standard, so these work in Claude Code, Codex CLI, and Cursor from the same folder. Some are original; many draw on prior art from the community, credited in CREDITS.md.

Skills are grouped by domain; workflows/ holds playbooks that compose atomic skills into an end-to-end process; meta/ holds domain-agnostic skills. The top-level domain is the install unit — install.sh installs only the domains listed in install.conf (so a non-coding domain never leaks into a coding agent). Within a domain, skills can nest at any depth — that deeper grouping is for humans — but leaf folder names must stay unique across the repo.

.
├── install.sh
├── <domain>/                 # atomic skills (e.g. software-development/)
│   └── <group>/<skill>/SKILL.md
├── workflows/                # playbooks that compose atomic skills
│   └── <workflow>/SKILL.md
└── meta/                     # domain-agnostic skills
    └── <skill>/SKILL.md

Each area documents its own skills — browse:

• software-development/ — design, planning, review, engineering
• workflows/ — multi-skill playbooks
• meta/ — domain-agnostic skills

Each skill's own SKILL.md is the single source of truth for what it does. This README stays deliberately thin so it doesn't rot as the collection grows.

./install.sh            # link skills from the configured domains into every supported agent

SKILL.md is a cross-agent standard, but each agent reads a different user-level directory. The union of two dirs covers all three (paths verified 2026-05-30 against the Claude Code, Codex, and Cursor docs — recheck if those tools move):

The installer links into both by default — every target dir is read by at least one agent, with no dead links. Edit a skill in this repo and every agent sees the change. Once linked, each skill triggers automatically by its description:, or is invocable as /<name>.

./install.sh --claude     # only ~/.claude/skills  (Claude Code)
./install.sh --codex      # only ~/.agents/skills  (Codex CLI)
./install.sh --cursor     # only ~/.agents/skills  (Cursor reads it)
./install.sh --copy       # copy instead of symlink (changes won't sync back)
./install.sh --force      # overwrite an existing foreign skill of the same name
./install.sh --uninstall  # remove only the links pointing back into this repo
./install.sh --dry-run    # preview, change nothing

What installs: only the top-level domains listed in install.conf (default-deny), so a new non-coding domain won't leak into your coding agents until you opt it in. Within each, every leaf folder with a SKILL.md is linked and the leaf name becomes the skill's command name, so leaf names must be unique repo-wide (the installer aborts on a collision). The installer refuses to overwrite a same-named skill it didn't create — pass --force to override.

(If an agent doesn't pick up a symlinked skill, re-run with --copy.) To publish the whole set as one installable Claude Code plugin later, add a .claude-plugin/plugin.json + marketplace entry; the folders are already shaped for it.

1. Copy docs/skill-template.md into a kebab-case folder in the right place — an atomic skill under a domain (<domain>/<group>/<name>/), a playbook under workflows/<name>/, a domain-agnostic one under meta/<name>/. Or run the skill-creator skill (/skill-creator) for a guided flow.
2. Add a one-line entry to that area's README.md.

Original skills are MIT (LICENSE). Sources for adapted skills are in CREDITS.md.