coderail-diagnoselisted

# CodeRail Diagnose Make the exact reported symptom reproducible before forming a root-cause theory. ## CodeRail boundary - Read `AGENTS.md`, `docs/NORTH_STAR.md`, `docs/TASKS.md`, and `git status`. - Work through the active CodeRail task. If none exists, identify the likely scope and test command, then use the public `coderail start` command before writing a reproduction. - Do not write `.coderail/tasks.json` or edit task state behind the CLI. - Never stage or commit task changes. - Finish through `python .coderail/coderail.py done`. ## Workflow 1. Restate the user's exact symptom as a pass/fail assertion. 2. Choose the highest public seam that can observe it. 3. Identify one fast command that will exercise that seam. 4. Ensure a CodeRail bug task owns the reproduction and likely fix paths: ```bash python .coderail/coderail.py start "Reproduce and fix <symptom>" \ --type bug \ --files "<production scope>" \ --files "<test or harness path>" \ --tests "<test or harness path>" \ --verify "<single reproduction command>" \ --accept "the exact reported symptom is covered by a regression check" ``` 5. Build and run the reproduction. Prefer, in order: - a focused automated test; - a CLI or HTTP fixture; - a headless browser assertion; - a captured-input replay; - a small throwaway harness; - a seeded property, stress, bisection, or differential loop. 6. Tighten it until it is red-capable, deterministic, fast, and agent-runnable. 7. Only then