My skills for Claude Code.
agents-kit/
├── CLAUDE.md # Core rules (behavior overrides)
├── skills/ # Shared skills
├── setup.sh # Creates symlinks
└── README.md
After running setup.sh, symlinks are created:
~/.claude/CLAUDE.md -> <repo>/CLAUDE.md
~/.claude/skills/_typescript/ -> <repo>/skills/_typescript/
~/.claude/skills/code-review/ -> <repo>/skills/code-review/
... # one symlink per skill
The kit has two parts:
CLAUDE.md— Behavioral overrides that aren't part of Claude's defaults (scope matching, git constraints, dependency rules). Loaded into every conversation.skills/— Each skill is symlinked individually into~/.claude/skills/. This lets you keep your own skills alongside the ones from this repo. Skills prefixed with_are auto-applied by convention (seeCLAUDE.md).
Note: If a skill with the same name already exists in ~/.claude/skills/, setup.sh will ask before overwriting.
The skills follow a natural development workflow:
- Understand —
/explore— Explore how something works — code, libraries, APIs, concepts, or architecture. Also use before planning to surface constraints, blast radius, and known alternatives. - Plan —
/designor/refactor— Break down a task into steps, compare approaches, and define scope before coding - Build —
/implement— Write production-quality code following project patterns - Review —
/code-review— Check for bugs, impact, and adherence to project patterns - Verify —
/verify— Verify a reported issue is real, identify root cause, and assess severity
You don't need to use every step — pick whichever skills fit the task at hand.
git clone git@github.com:drimchansky/agents-kit.git
~/agents-kit/setup.shThe repo can be cloned anywhere — setup.sh resolves its own location automatically.
To update later:
cd ~/agents-kit && git pullNo need to re-run setup.sh — symlinks point to the repo, so pulling updates the skills in place.