codify-learningslisted

# Codify Learnings Turn recurring in-session discoveries into compounding, reusable assets. This skill only *drafts* candidates into the approval queue — nothing is written to a memory file, skill, or CLAUDE.md until a human approves the candidate via `/engineer-agent review-queue` (which delegates the write to `execute-item`). ## Tools Needed - `Read` — read config and queue items - `Glob` — find queue items by date - `Grep` — scan item bodies for friction signals - `Bash` — date math, aggregation - `Write` — write candidate drafts ## Input An `/engineer-agent codify` invocation, optionally with `--since <date>|last-week` (default: last 7 days) and `--project <slug>`. ## Steps ### 1. Load Config Read `~/.local/share/engineer-agent/engineer.yaml`. Extract the `projects` map (each entry's `path` and slug). If missing, stop and tell the user to run `/engineer-agent setup`. ### 2. Scan Recent Work Determine the cutoff date from `--since` (default: 7 days ago). Using the same date-scan approach as `generate-digest`, glob `~/.local/share/engineer-agent/queue/completed/` and `~/.local/share/engineer-agent/queue/rejected/` for items created since the cutoff (filter by `--project` if given). For each item, read the body — especially the `### Findings & Disposition` ledger, `### Remaining Work`, rejection reasons, and any debugging or environment notes. ### 3. Cluster into Candidate Learnings Look for **recurring** signals worth codifying — the value is in things seen more