playwright-test-diagnosis

## Goal Turn `heal-traces.ndjson` into a concise root-cause report with statement-level evidence. ## Inputs To Locate - `heal-traces/<executionId>/<playwrightTestId>/<attempt>/heal-traces.ndjson` - Sibling artefacts in the same `<attempt>/` dir: `trace.zip`, `screenshots/stmt-*.png`, `videos/<file>.webm` (when present) - `heal-traces/executions.ndjson` — append-only run index; the latest line is the most recent run - `heal-traces/<executionId>/execution.json` — per-run manifest (totals, git, duration, per-test summaries) - The test file mentioned in the trace (`statement.loc` and source text) If multiple trace files exist, prioritize the latest failing attempt. `<executionId>` is an auto-generated uuidv4 per run; pick the newest entry in `executions.ndjson` to find the most recent execution dir. ## Workflow 1. Find the `test-result` event and note final status, duration, stderr, and stdout. 2. Scan statement events in order and identify: - first failing statement (`status != ok`) - last successful statement before failure - slow statements (`durationMs` outliers) 3. Extract high-signal fields from failing/nearby statements: - `source` - `loc.line` and file context - captured variable values - serialized error payload - screenshot references (if present) 4. Build a timeline: - setup/context - trigger statement - error manifestation - propagation/teardown 5. Correlate with external signals: - network/API response mismatches ...