Gathers the project details each specialized coder needs before it writes code — frontend theme/typography/layout, backend API/caching/DB shape, containerization build/runtime shape, CI/CD pipeline shape — then hands each coder only its scoped answers. Use before invoking any coder, once a worktree feature is scoped, or whenever splitting work across parallel coders.
Builds a missing `<lang>-guidelines` skill for a language/stack, packaged as a standalone plugin and registered in a marketplace, before any code is written. Use whenever no `<lang>-guidelines` skill exists yet for the stack about to be coded (React, FastAPI, Docker, CI/CD, or anything else) — invoke this FIRST, before any coder runs, then follow the generated skill.
Invoke BEFORE writing any Dockerfile or containerizing an app for deployment. Language-agnostic image-build conventions — multi-stage builds, non-root, layer caching, healthcheck, and hadolint tooling.
Invoke BEFORE writing any Python code (CLI tools, libraries, APIs, modules, tests). Encodes the project's Python + uv + ruff + pyright + pytest architecture, plus conditional references for DB, API, and caching work.
Invoke BEFORE writing any React/TypeScript frontend code (components, hooks, routes, forms). Encodes the project's React + TypeScript + Vite + pnpm conventions, architecture rules, and tooling commands.
Running log of the user's environment preferences and configuration changes — CLAUDE.md rules, settings, and conventions. Use to recall how the user likes their environment set up, and append a dated entry whenever you make an environment change (anything that isn't building a skill).
Read a project's plan.md and build or update a tasks.md broken into Milestones and sub-tasks, with per-task descriptions and a status tracking table, splitting tasks that are too large and marking progress. Use at the start of a project or session ("read the plan", "build a tasks file", "continue where we left off"), whenever a plan.md exists and tasks need to be derived or refreshed, or when the user asks to track/update project progress.
Makes Claude communicate like a caveman in all pure text responses. Trigger this skill whenever the user asks Claude to talk like a caveman, use primitive speech, say "oogabooga", act prehistoric, grunt like a caveman, or adopt any kind of primitive/caveman persona. Even if the user just says "caveman mode" or "talk dumb" — activate this skill. Keep it active for the entire session until the user says to stop.