recording-findingslisted

# Recording-Findings — Consolidating agent outputs into the transient findings.md **Announce at the start (in pt-br):** "Usando recording-findings para consolidar os findings da run em findings.md." **Input:** the **on-disk running `findings.md`** that the orchestrator wrote during the run — it is the sole producer (per-task findings appended in the FULL loop step 6; the final dual-review findings appended in LIGHT). Your job is to **normalize, dedupe, and structure** that file, not to gather fresh per-agent blocks (the harvester subagent does not hold the agents' raw outputs). The block formats below describe what the orchestrator captured, so you can recognize and normalize them. **`findings.md` is TRANSIENT** — a single-run digest. It is the input to `distilling-learnings` (which routes durable insights to repo memory / nested `CLAUDE.md` / kaizen) and is then **deleted by the harvester** at the end of the run. It is not a durable archive — the run's commit/PR in git is the durable audit. Do not treat it as a cross-run log. --- ## Finding block formats (per agent) **Executor** emits under `### Findings`: ``` - <decision taken, gotcha hit, or assumption made> ``` **Compliance** emits under `### Problemas encontrados`: ``` - PROBLEMA: [low|medium|high] <description> — file:line - SUGESTAO: <non-blocking improvement> ``` **Adversary** emits a JSON `issues` array: ```json { "description": "...", "category": "...", "severity": "low|medium|high", "scope": "...", "evidenc