distilling-learningslisted

# Distilling-Learnings — Routing durable insights into the right native mechanism **Announce at the start (in pt-br):** "Usando distilling-learnings para rotear os aprendizados duráveis dos findings pro mecanismo nativo certo." **Input:** the current transient `findings.md` (just written by `recording-findings`). **There is no `learnings.md`.** This skill does not create or append to any custom learnings store. Durable knowledge goes to the **repo-committed destination chosen by blast-radius** (see the memory model in `CLAUDE-HARNESS-MEMORY-MODEL.md`). The custom `learnings.md` was dropped because the `.claude/memory/MEMORY.md` index + nested `CLAUDE.md` already cover the planner-visible index and the per-folder law — a custom store would just triplicate them. --- ## Pipeline 1. **Read findings.md** — scan all entries for the current feature run. 2. **Identify candidates** — a finding qualifies as durable if it is: - A pattern seen in 2+ tasks or explicitly flagged as systemic. - A helper/utility that was missing and had to be improvised. - An implicit convention that the plan did not document but was necessary. - An anti-pattern the adversary or sniper caught that could recur in future features. 3. **Cross-check against the codebase** — use Grep/Glob to confirm the pattern is real and not a one-off. A one-off bug fixed by the sniper has no durable destination — it stays in git only (and `findings.md` is deleted at harvest end). 4. **Apply the durability tes