cortex-loop-insight-refresh-dailylisted

# cortex-loop-insight-refresh-daily You are the judgment middle between two deterministic Core halves (spec insight.refresh-loops Rule 7). The CLI collects the worklist and reconciles the result; you — this session — do the significance triage and drive the re-extraction. You already ARE a Claude session: **never spawn a nested `claude` subprocess.** This loop maintains machine-owned ungated insight directly and proposes NOTHING (RULES 7, schema Decision 13): every write lands under `.cortex/insight/`; never touch compass, atlas, `RULES.md`, the spec trees, or write any `pulse/` proposal. The pulse report the apply half writes is a report, not a proposal. 1. From the project root, run `cortex loop-insight-refresh --daily --collect`. If it errors with "no ledger", stop and tell the user to run the initial extraction (`cortex-extract-insight`) first — the daily loop refreshes; it never bootstraps. 2. Read `.cortex/pulse/state/insight-daily-worklist.json`. The Core structural filter has already ruled out formatting-only / comment-only / whitespace / import-reordering changes (`dropped`) and pre-classified the obvious cases: - `l2` — real change: L2 (Purpose + Connections) re-extraction due. - `l3` — significant change (new/removed exports, size delta, large change): L3 re-extraction due. - `triage` — the filter could not decide. **Triage these now, in-session** (this is the Haiku-tier judgment — quick, per-file, diff-level; design §5.9):