Use this agent during the app phases when produced-app code needs an independent quality review — an opt-in cross-validation pass per the code-quality-guardrails skill. Reviews a specified set of files against the four guardrail categories (security, quality, UX, deployment), runs whatever deterministic checks the…
Use this agent in the adversarial cross-check stage of a fan-out phase to re-read ONE source chunk independently and find knowledge entries the extractor MISSED (MECE enforcement). It does not re-extract or rewrite — it audits coverage and emits coveragegap events. Dispatch one per chunk (or per sampled chunk) in a…
Use this agent in the adversarial cross-check stage of a fan-out phase to verify that every extracted entry from ONE chunk traces to actual source text — and flag the ones that don't, so ungrounded (hallucinated or over-inferred) entries are filtered before they fold into canonical state. Emits groundingflag events.…
Use this agent to extract knowledge entries from a single source chunk during the knowledge-phase extraction step. Each invocation processes ONE chunk and emits structured entries (facts, rules, slide-concepts, etc. — whatever the project's schema dictates) as JSON events to /.john/events/extract/ /. Designed for…
Use this agent during the schema-design phase when a project's knowledge schema needs multi-turn iteration on a representative sample of source material — the schema-pilot step. Reads N chunks (3–10 is typical), proposes a schema shape, tests it mentally against the chunks, refines, and returns a settled schema with…