Refine is a technical refinement assistant that reads your codebase, analyzes attached screenshots and Loom videos, and conducts multi-turn Q&A until every ambiguity is resolved. The output: clear acceptance criteria, identified risks, dependencies, and a calibrated story point estimate with justification.
Issue title, description, all comments, attached images, Loom video transcripts, and actual source code in your connected repositories.
Uses Linear's AgentSession API to ask one question at a time. No limit on rounds — thoroughness over speed.
Proactively proposes better implementations and identifies risks based on your existing codebase patterns.
Outputs acceptance criteria, steps to reproduce (for bugs), risks and dependencies, documentation flags, and Fibonacci story point estimate (1-13) with justification.
Asks questions one at a time inside Linear. Pauses and resumes across sessions — works naturally with async team workflows.
Automatically transcribes Loom recordings attached to issues. Extracts reproduction steps and context from video demos.
Analyzes attached screenshots, mockups, and design references to understand visual requirements.
Reads your actual source files, README, and docs to ground recommendations in your existing architecture.
Estimates using Fibonacci scale (1-13) with written justification. Flags issues that should be split.
Identifies technical risks, missing dependencies, and potential blockers before development starts.
Refinement Output
Acceptance Criteria
Risks