reasoning-trace-recorder

> **AI-consumed reference.** Optimized for Claude to read during execution. > Human-readable explanation: see [docs/architecture/HIERARCHICAL_PLANNING.md](../../../docs/architecture/HIERARCHICAL_PLANNING.md) > or [docs/getting-started/](../../../docs/getting-started/) depending on topic. # Reasoning Trace Recorder **STATUS — v3.7.0-alpha.2 (Milestone B).** Forensic reproducibility per spec goal §2.1.3. ## Behavior 1. Detect: if `.claude/plans/active.json` does NOT exist → exit silently 2. Read `.claude/plans/active.json` — if `active.task` is null → exit silently 3. For every emitted event from companion hooks/tools, append a line to `.claude/plans/traces/{TASK_ID}.jsonl` ## Event schema (one JSON per line) ```json { "ts": "2026-04-29T14:32:01Z", "event_id": "TR-00101-007", "task_id": "TASK-00101", "type": "output_claim", "payload": { "claim": "src/auth/jwt.ts exports verifyToken", "grounded": true, "grounded_by": ["TR-00101-003"] } } ``` ## Event types | Type | Emitted by | Required payload fields | |------|-----------|-------------------------| | `file_read` | Read tool / hook | `path`, `lines_read`, `sha256` | | `output_claim` | Claude reasoning | `claim`, `grounded` (bool), `grounded_by` (array of event_ids) | | `tool_call` | PreToolUse hook | `tool_name`, `args_hash` | | `tool_result` | PostToolUse hook | `tool_name`, `exit_code`, `result_hash`, `duration_ms` | | `decision` | master-planner | `decision`, `reasoning`, `confidence` | | `pha...