llm-dark-patterns

Out-of-band your AI Code Stop / SubagentStop hook judges that block 30 LLM dark patterns at the closeout boundary: false-success without evidence, sycophancy, paternalism, permission-loops, vibe time estimates, fake recall, fake stats, fake citations, count-vs-enumeration self-consistency drift, context loss after compaction, multi-agent rollup failures, power-user polish failures, and unreachable new symbols (advisory).

Deterministic verdicts, no network calls, no model in the verdict path