KB Review

Run the canonical governance lane between draft knowledge and the approved live brain. kb-review owns two internal modes:

• gate mode for pending draft decisions, promotion/rejection, and briefing rebuilds tied to the same review pass
• maintenance mode for approved-layer drift, stale briefings, provenance and alias refresh, creator consistency checks, governance-index rebuilds, graph gaps, backlog pressure, and safe mechanical fixes

In Karpathy's LLM Wiki, this maps to two operations: the implicit quality judgment during ingest (our gate mode), and the explicit "Lint" pass he describes — "look for contradictions between pages, stale claims that newer sources have superseded, orphan pages with no inbound links, important concepts mentioned but lacking their own page, missing cross-references" (our maintenance mode). This contract makes both operations explicit and separable.

When a user says output/reports, decide whether they mean a governance report or a deterministic rendered report. Lint, contradiction, orphan-page, and stale-claim work stays here in maintenance mode.

Minimal loop

1. inspect the draft package, cited captures, and overlapping live pages
2. decide approve, reject, or escalate for human judgment
3. promote only durable, well-supported knowledge into live
4. rebuild briefings when the approved change should alter downstream context

When this compounds the wiki