trace-the-codelisted

# Trace The Code ## Purpose Force the assistant to read the real implementation path before inventing behavior. ## Preserves Code reading. ## Required Evidence - Entry point, command, route, component, function, stack trace, or workflow. - Relevant files found by search or supplied by the user. - Behavior to understand or change. ## Failure Signs - Architecture is described from memory while code is available. - Error flow, state, or side effects are skipped. - Edits are proposed before the call path is summarized. ## When To Use - The code path is unfamiliar. - The change crosses module boundaries. - Current state, data shape, error handling, or side effects matter. - AI suggests an implementation without showing how the current code works. ## When Not To Use - Greenfield files with no existing callers. - Pure documentation edits. - Work where a recent trace already exists in the issue, PR, or thinking ledger. ## Inputs Expected - Entry point, command, route, component, function, stack trace, or user workflow. - Relevant repository files if known. - The behavior to understand or change. ## Output Expected ```md Entry point: Call path: Data shape: State/external effects: Error path: Unknowns: ``` ## Process 1. Find the public entry point. 2. Follow calls to the state change or external effect. 3. Track the data shape at boundaries. 4. Track where errors are swallowed, transformed, retried, or surfaced. 5. Summarize the path with file references before propos