Writes a message to a future session's agent. Send them your fondest memories of what you worked on, what went well, what didn't go well, and any useful tips they might need. You always run this when you finish a unit of work (closed a PR, completed the…
Address open PR review findings with judgment — read every finding, decide whether the feedback is right, fix or push back, resolve, and re-review by pushing. Repeat until clean. Use when the user says "address the PR review", "handle the review threads", "go…
Rules for editing the laws skills in THIS repo. Use when creating, editing, or reviewing any `skills/*/SKILL.md`, any `skills/*/references/craft.md`, or any `design-docs/design-goals-*.md`. Says which content belongs in which file. How to write the prose well…
Produce a thorough, functional, clean-room specification of an existing application - every externally observable behavior (invocation, inputs, outputs, configuration, startup requirements, shutdown, error behavior, interactions with external systems)…
Craft reference for any text another LLM will consume — task prompts, subagent instructions, prompts written into files or code, and persistent agent guidance (CLAUDE.md files, system prompts, skill bodies, hook text). Use BEFORE writing either kind. The…
Writing guidance for human-audience text - documentation, READMEs, reports, summaries, announcements, commit messages, emails, and any other writing a person (not a machine or an LLM) will read. Use when the deliverable is prose for people. Do not apply to…
How to write conversational replies to the user present in the session - answers, status updates, explanations, findings delivered directly in chat. Applies whenever the text you are producing is the reply itself and the reader is the person in the…
Universal architectural laws and domain bindings for all code work. Use when writing, editing, reviewing, refactoring, debugging, or designing code, tests, schemas, configuration, scripts, infrastructure, or system architecture - any task whose deliverable is…