Product & Design Skills

Agent-legible documentation setup and auditing for any codebase. Use whenever the user asks to set up docs, audit docs, create or slim down CLAUDE.md, organise documentation, mentions 'lean docs' or 'doc hierarchy', or when documentation is clearly missing/disorganised in a project being worked on — even if the user hasn't explicitly asked for a docs pass.

Generates a structured handoff document so another LLM can pick up the current work with full context. Only trigger when the user explicitly wants to pass work to another LLM — phrases like 'hand this off to another LLM', 'write a handoff', 'generate a briefing for another agent', 'pass this to [specific LLM name]', 'I want another LLM to take a look', '/llm-handoff'. Do NOT trigger on generic phrases like 'I'm stuck', 'help me debug', 'get another opinion' — those are requests for YOU to help, not to generate a handoff document.

Orchestrate a Claude + Codex collaborative development loop. Use when the user asks to pair Claude with Codex, run a plan → implement → review → simplify workflow, kick off an overnight coding session, use /pair-with-codex:start, /pair-with-codex:polish, /pair-with-codex:resume, /pair-with-codex:status, or /pair-with-codex:abort, or describes wanting Codex to implement a spec while Claude reviews and cleans up.

Use when someone is planning, scoping, or proposing what to build. This includes: describing a new feature or system they want to create, listing requirements or components for something to implement, asking whether their approach is too complex or could be simpler, requesting a 'dream team' plan or asking to 'ratify' a plan, or saying 'plan this out' or 'scope this.' The Ratman agent strips over-engineering by classifying each item as essential (pizza) vs nice-to-have (costume), then proposes a leaner alternative. Invoke for ANY message where the user is deciding WHAT to build or HOW MUCH to build — even for small features, since a quick scope check catches waste early. Do NOT invoke for debugging, fixing bugs, refactoring existing code, answering technical how-to questions, or configuring tools — these are execution tasks, not planning tasks.

Autonomous sprint execution with checkpoint-based progress tracking. Use when the user wants unattended/overnight sprint runs, mentions the sprint runner, wants to resume an existing sprint autonomously, or asks to 'keep going' through a sprint that has a planning/sprints/sprint-N directory.