Operations
Social
Commands → Skills
Claude Code merged custom commands into skills, and kun followed: every verb that used to
live in .claude/commands/*.md is now a skill at .claude/skills/<name>/SKILL.md. Same
invocation (/qa admission still works), same bodies — plus what skills add:
| Gain | What it means |
|---|---|
when_to_use routing | Each skill's frontmatter carries its natural-language triggers, so prose like "qa admission" or "is it green" activates the right verb without a slash — passive activation is now a per-skill contract, not one giant table. |
| Supporting files | A skill is a directory: SKILL.md + scripts/ + references/ (see clone). |
| Progressive disclosure | Only name + description sit in context until a skill fires. |
| Invocation control | context: fork runs heavy skills in a subagent (see watch); user-invocable / disable-model-invocation tune who can fire what. |
Where everything went
| Was | Is |
|---|---|
.claude/commands/<verb>.md | .claude/skills/<verb>/SKILL.md |
~/.claude/commands/ copies | pruned by setup.sh (a same-named skill shadows a command anyway) |
| This page's command tables | Skills + the keyword registry |
The full verb inventory lives in the vocabulary registry (.claude/vocabulary.json,
browsable at keywords); per-verb specs are the skill files themselves.
/health verifies the skill fleet against engine.json counts.