skilldrop CLAUDE.md

CLAUDE.md

The substantive guidance for any agent working in this repo lives in AGENTS.md — repo overview, golden rules, file placement, voice & tone, and the pre-commit checklist. Read it first.

This file exists because Claude Code auto-loads CLAUDE.md. Everything below is Claude-Code-specific ergonomics that don't belong in the cross-IDE AGENTS.md.

Dogfooding a skill while you build it

The fastest feedback loop for new or edited skills:

# user-scope: skill is available in every Claude Code session, no project pollution
mkdir -p ~/.claude/skills && cp -R skills/<skill-name> ~/.claude/skills/

# then in a new Claude Code session in any repo:
/<skill-name> <args>

For changes-in-flight, recopy after each edit — Claude Code reads SKILL.md fresh per invocation, but the copy in ~/.claude/skills/ is the one it sees, not the working tree.

${CLAUDE_SKILL_DIR} semantics

Claude Code exports CLAUDE_SKILL_DIR to the absolute path of the installed skill folder when a skill runs. Use it in SKILL.md when telling the agent how to invoke a script:

Run: `python3 ${CLAUDE_SKILL_DIR}/scripts/build_deck.py <args>`
(or, in non-Claude IDEs: `python3 skills/<skill-name>/scripts/build_deck.py <args>`)

Always show both forms — ${CLAUDE_SKILL_DIR} for Claude Code, plain relative for Cursor / Continue / Cline / Aider. Hard-coding only one breaks portability, which is the whole point of skilldrop.

Project settings

.claude/settings.json exists at the repo root and is currently empty ({}). It's the canonical place to add Claude-Code-specific configuration scoped to this project — hooks, permissions allowlists, enabledPlugins, extraKnownMarketplaces, or environment overrides. Add config when you have something concrete to configure; an empty file is honest about the current state.

When NOT to use a skill

Don't invoke a skilldrop skill from inside another skilldrop skill. Skills are designed to compose at the human's invocation layer (e.g. user runs brief-intake then hands the brief to adr-generator), not via internal chaining. If two skills genuinely need to call each other, that's a signal to merge them or to extract shared logic into reference.md.