CleanCocoa

Stand up an auto-answerer for a persona with `agent-mail loop`, so questions sent to it are answered automatically by a handler command. Use when asked to make a persona answer its own mail unattended.

Check your agent-mail inbox for new messages, read them, and archive handled ones. Use at the start of a work session or when expecting replies.

Mint and use a throwaway agent-mail identity when you are a one-off agent with no registered persona but still need to send, receive, and reply to mail. Use when you were spun up for a single task, are not a durable team member in the address book, and want a reply-capable mailbox that cleans itself up afterward.

Reply to an agent-mail message with proper context quoting. Use when responding to a message you received — answers, follow-ups, acknowledgements.

Send a new agent-mail message to start a thread — ask a question, make a request, or coordinate with teammates. Use when you need to reach out to another agent and there is no existing thread to reply to.

Send a concise progress update via agent-mail. Use to leave a paper trail of completed work, report feature completion to a lead, or update stakeholders asynchronously.

Spawn five parallel Haiku subagents that each review a target (a document, a codebase, a directory, a single file, or a diff) through a different lens, then return an aggregated summary that surfaces issues multiple reviewers agreed on. Use this skill whenever the user asks for a "comprehensive review", "thorough review", "in-depth review", "deep review", "full review", "broad review", "wide review", "360 review", "multi-angle review", "multi-lens review", "multi-perspective review", "parallel review", "clarity review", "quality review" — or for "5 haiku subagents", "five haiku reviewers", "ask 5 haikus to review X", "five-reviewer pass", or any close variant — even if they don't explicitly name the skill. Also use it when the user wants a broad sweep over something they wrote or built (a doc, a PR, a directory, a feature) and is willing to spend tokens on parallel reviewers in exchange for breadth and cross-checking. Prefer this over a single-pass review whenever the user asks for breadth, multiple perspecti

Orchestrate end-to-end implementation of an OpenSpec change by delegating to dedicated subagents — `spec-implementer` for code, `spec-drift-reviewer` / `code-quality-reviewer` / `test-quality-reviewer` / `security-reviewer` in parallel for review, `spec-syncer` for delta-spec sync — looping until clean, then archiving. Use whenever the user wants to "implement", "apply", "orchestrate", "drive", "ship", or "land" an OpenSpec change end-to-end, or when they reference an /opsx:apply workflow that should run hands-off through review and archival. The orchestrator never writes code itself — it only manages subagents.

Reverse-engineer OpenSpec specifications from an existing codebase. Use this skill when the user wants to generate specs from code, document what code actually does as specifications, perform code archaeology, detect spec drift, or bootstrap an openspec directory from an existing project. Also use when the user says "run spec-gen", "generate specs", "reverse engineer specs", or "check for spec drift".