Skills, in their native habitat.

Use to cut AI code slop, the over-building, boilerplate, one-use abstractions, needless dependencies, and 50-lines-where-1-would-do that the AI writes by default. The code-side...

Use to surface Codex features the user is not using but should, tuned to the friction they describe. Trigger when the user says "what can Codex do", "is there a better way", "I...

The first-run setup wizard, played as a live four-act "opening night" for the kit. It proves the engine is alive (staggered doctor cascade), equips the agent with the essential...

Use to get an honest, independent design review of a screen, does it look good, or does it look AI-made? It looks at a real screenshot, scores it, and lists the fixes. It REPORTS,...

Use when something broke badly, work seems lost, a merge went wrong, or git is scary, to calmly get back to a working state without making it worse. Trigger when the user says...

Use to PROVE a change actually works before calling it done, by really running or opening it and reporting what you saw. Run after /step. Trigger when the user says "/check",...

Use before anything goes public or gets deployed, to catch leaked API keys, exposed .env files, plain-text passwords, secrets in frontend code, and the obvious security holes....

Use to build ONE step of the plan at a time, safely, with a checkpoint after it works. Run it after /start has produced an approved plan and a CLAUDE.md. Trigger when the user...

Use to surface the Claude Code features the user isn't using but should, tuned to what they're actually doing or struggling with. Trigger when the user says "/claude-tips", "what...

Use when something is broken or behaving wrong, to find the REAL cause before changing anything, instead of piling on patches. Trigger when the user says "/fix", "it's broken",...

Use to give your agent real outside tools via MCP, web crawling, web search, live docs, a browser, instead of guessing from stale memory. Works in Claude Code, Codex, and...

Use to get a plain-English readout of what the AI actually built, so you understand it and could defend it, the shape and the important parts, never line-by-line. Trigger when the...

Use before you accept a change, to look at the actual diff with fresh eyes against the original goal and catch what's risky, half-done, or off-scope. Trigger when the user says...

Use to get the project LIVE on the internet, deploy it, set production env vars, and hand back a working URL. Trigger when the user says "/ship", "deploy", "put it online", "make...

Use BEFORE designing anything, to interview the user and write a DESIGN.md (the design source of truth) so the app has a real aesthetic instead of a default AI look. Trigger when...

Use at the very START of any new app, feature, or change, BEFORE writing any code. The assistant interrogates the user until the goal is genuinely clear, then writes a short plan...

Use to FIX the things that make a screen look AI-made, working from DESIGN.md and the anti-slop catalog, fixing by root cause not by decoration. Trigger when the user says...