Routing¶
The toolkit detects intent from plain English and routes to the right skill automatically. You don't need slash commands for any of these.
User-triggered routing¶
| You say | What runs |
|---|---|
"add [X]", "implement [X]", "build [X]" |
/implement — full 5-phase workflow |
"fix [X]", "debug [X]", "not working" |
superpowers:systematic-debugging |
"check this", "review [X]", "before PR" |
/standards-check |
"new project", "starting fresh", "scaffold this" |
/new-project |
"switch project", "change project" |
/project |
"push to git", "push this", "open a PR" |
/git-push |
"upgrade toolbox", "update toolbox", "/upgrade" |
/upgrade |
"set vault", "change vault path" |
/set-vault |
"delegate to codex", "use codex for" |
/codex-delegate |
"review with codex", "codex review" |
/codex-review |
"update docs", "docs are stale", "/update-docs" |
/update-docs |
Automatic routing (no user prompt needed)¶
These fire based on Claude's own internal actions — no user message required:
| Claude is about to… | Skill that fires first |
|---|---|
Call WebFetch or open a URL |
/web-fetch — routes through Crawl4AI |
| Run multiple git commands | /git-ctx |
Read a git diff to draft a commit or PR |
/diff-summary |
| Read one section from a file > 100 lines | /read-section |
| Look up an npm or PyPI package | /pkg-info |
Check runtime versions, ports, Docker, or .env |
/env-check |
| Search code with Grep | /grep |
| Write to any memory file | /memory-sync |
How routing works internally¶
The routing table lives in ~/.claude/CLAUDE.global.md. Each row is a phrase pattern
and a skill file path. When a user message matches, the skill is read and followed before
any response. See Contributing / Adding a Skill to
add a new routing row.