claude-code-lab

Watch Anthropic's shipping surface (CHANGELOG, README, code.claude.com docs) and propose lab-shaped edits as a pull request. The data plane (scripts/freshness/run.ts) is the only source of truth — every claim in every patch must trace verbatim to the delta file. Reads the full lab content (10 spine chapters + library entries + design system) and decides per-item whether something released by Anthropic should update specific lab content. Composes EN + CS edits, voice-judge sub-agents enforce both provenance (no fabricated version numbers, URLs, or feature claims) and voice match. Branches, commits, opens a PR via gh CLI. Never merges. Never synthesizes a delta. Activate when the user says "run freshness", "freshness check", "what's new from Anthropic", or invokes /cc-lab-freshness.

Diagnose a Claude Code setup against cc-lab patterns. Two modes — project (would a teammate cloning this repo succeed today?) or user (is your personal harness well-tuned?). When invoked without an explicit mode arg or scope keyword, asks via AskUserQuestion before reading any files. Returns 3-5 evidence-grounded observations grounded in cc-lab chapter knowledge, with copy-paste artifacts and chapter links. Activate when the user says "diagnose my setup", "review my CLAUDE.md", "audit this repo", or invokes /cc-lab-diagnose [project|user|both].

Capture desktop and mobile screenshots for cc-lab chapters with consistent state (dark mode, 2× retina, no PII, no notifications) using the computer-use MCP. Activate when the user asks to "take a screenshot for the lab", "capture for cc-lab", "shoot the voice chapter visuals", or hands you a shot list JSON/YAML.

Companion skill for the claude-code-lab workshop. Walks the user through the first four chapters interactively — install, first task, CLAUDE.md, iteration. Use when the user wants guided practice with Claude Code, mentions "cc-lab", "claude code lab", or asks to walk through the workshop.