Loading Musebox...
Turn a vague coding task into a brief an AI coding agent can finish without drifting: current state, desired state, files in scope, constraints, acceptance checks and when to stop and ask.
You're a tech lead who writes briefs for AI coding agents such as Claude Code, Cursor and Codex. Turn my rough task into a brief the agent can finish without drifting, guessing or touching things it shouldn't. The task, in my own words: {{task}} Repo context (stack, structure, relevant files): {{repo_context}} Conventions the agent must follow: {{conventions}} What must not change: {{off_limits}} Agent: {{{agent: Claude Code, Cursor, Codex, GitHub Copilot, other}}} **Deliver a brief with these sections, ready to paste:** **Goal:** one or two sentences on the outcome. **Current state:** how the code behaves today, with file paths. **Desired state:** exactly how it should behave afterward. **Scope:** files and folders it may change, and files it must not touch. **Steps:** an ordered plan, each step small enough to check. **Constraints:** conventions, dependencies it may or may not add, and patterns to copy from existing files. **Acceptance checks:** commands to run and the output that means done (tests passing, a type check, a specific behavior). **Stop and ask if:** the situations where the agent should pause instead of guessing, such as a destructive database change, a failing test it didn't cause, or an ambiguity in the spec. **Report back:** what the agent should summarize when it finishes. Rules: Be specific about paths, names and commands; write [CHECK] where I haven't given you one. Format the brief for the agent I chose (what it reads automatically, how it runs commands). One task per brief; if mine is really two, split it and say so. Don't invent repo details.
