You MUST use this before any creative work — creating features, building components, adding functionality, or modifying behavior. Settles intent, requirements, and design through rounds of questions with recommended answers, writes a spec reviewed via…
Use when asked to implement a bean, work on a bean, or pick up a task by its bean id (e.g. "implement dotfiles-4byb", "work on this bean"). Drives the full loop — read, validate, branch, implement, subagent review, user review, then land via a PR with…
Use when you have a spec or requirements for a multi-step task, before touching code. Breaks the spec into tracer-bullet vertical-slice tickets — behaviour, acceptance criteria, and blocking edges, not implementation steps; emits an epic with task children…
Request a code review from the user via plannotator. Use at the end of an implementation task, or when the user asks for a review cycle. Collects structured annotations, then addresses each item.
Challenges AI-generated plans, code, designs, and decisions before you commit. Pairs with any other skill as a review layer. Uses pre-mortem analysis, inversion thinking, and Socratic questioning to find what AI missed — blind spots, hidden assumptions,…
Use when completing development phases or branches to identify and update CLAUDE.md or AGENTS.md files that may have become stale - analyzes what changed, determines affected contracts and documentation, and coordinates updates
Use when creating or updating CLAUDE.md files for projects or subdirectories - covers top-level vs domain-level organization, capturing architectural intent and contracts, and mandatory freshness dates
Conduct rigorous, adversarial code reviews. Use when users ask to "critically review" my code or a PR, "critique my code", "find issues in my code", or "what's wrong with this code". Identifies security holes, lazy patterns, edge case failures, and bad…