agentic-analyzer

Authoring kit for LLM-judged, schema-validated, replayable code analyzers. Scaffolds a domain-specific Claude Code skill via an interactive `/new-analyzer` interview that drafts rules.md through a dispatched rule-author subagent, auto-generates one fixture stub per rule, optionally seeds a dev-team expected-entities oracle, and gates success on a post-stamp validator. Stamped analyzers record per-entity property bags (Phase D side-effect + Phase D.5 consolidation) and consult the oracle to backstop discovery with within-run framework expansion. Ships five specialist subagents, a companion `/expected-entities` command, and a generic runtime (schema validator, scaffold validator, snippet normalizer, fixture comparator, oracle comparator, override-replay engine).