Skip to content

Code Tool Skills

These skills are triggered automatically by Claude's own internal actions — no user prompt required. They enforce efficient, low-waste tool use.

/read-section — extract one symbol from a large file

Auto-triggered: when Claude needs one function, class, or section from a file over 100 lines.

Runs tools/read-section/read_section.py with language-aware extraction:

python tools/read-section/read_section.py --file src/auth.ts --fn validateToken

Returns only the requested block — not the full file. Prevents loading a 400-line controller to read a 30-line function.

Auto-triggered: before any code search with Grep.

Enforces a two-phase pattern: 1. First pass: retrieve only matching file paths (no content) 2. Second pass: read content from the specific files identified

Broad content searches across the full repo on the first attempt are prohibited. All searches cap output with head_limit to prevent unbounded result sets.

/diff-summary — structured diff context

Auto-triggered: before reading git diff to draft a commit message or PR body.

Produces a structured summary of what changed and why — file-by-file, with a one-line description per change. Used by /git-push and /implement Phase 5.

/git-ctx — batch git state

Auto-triggered: when Claude needs to run multiple git commands (status, log, diff, branch).

Batches all git state into a single parallel tool call instead of sequential commands. Returns one structured context block covering branch, status, recent log, and diff stat.

/env-check — runtime environment

Auto-triggered: when Claude needs runtime versions, running ports, Docker state, or .env file presence.

Runs tools/env-check/env_check.py — checks Python/Node/Go versions, open ports, Docker container status, and whether .env exists and is git-ignored.

/pkg-info — package metadata

Auto-triggered: when Claude needs version, types, license, or popularity for an npm or PyPI package.

Runs tools/pkg-info/pkg_info.py — queries the registry and returns the latest version, TypeScript types availability, license, weekly downloads, and homepage.

/index-repo — build structural code index

Trigger: "/index-repo" (run once per project, re-run after major changes)

Builds a structural map of the codebase — every file, its exported symbols, imports, and call relationships. Stored in .claude/index/ as JSON files.

After indexing, structural questions ("what calls X?", "what imports Y?") are answered by a Haiku sub-agent that reads only the relevant index file — not the source. Keeps searches fast and context usage low.

Supports incremental re-runs (only changed files re-processed).

/web-fetch — cached clean web content

Auto-triggered: before any WebFetch call.

Routes fetches through the local Crawl4AI container (if running). Returns clean Markdown output and caches pages for 24 hours — the same URL is never fetched twice in a day. Falls back to Claude's built-in WebFetch if Crawl4AI is unreachable.