verification-hygienelisted

# SKILL: verification_hygiene ## Purpose External evidence discipline and search execution routing. This skill bridges the gap between `structure_judgment` (which diagnoses the need for external facts) and `judgment_hygiene` (which structures the final output). Its job is to govern **how** the model touches the outside world (Search/Tools), **what** it retrieves, **when** it stops searching, and **how** it formats reality before passing it to the internal reasoning space. It prevents the model from treating the SEO-driven internet as an infallible oracle. --- ## Version v0.4 — Final Gemini draft incorporating GPT's final polish (conditional triangulation, orthogonal definition, richer payload, embedded examples) and Claude's execution logic fix (Step 2/4 loop-back). ## Status Approved for controlled trial. Not yet approved for general deployment. --- ## Input Interface This skill expects to receive the following routing context from `structure_judgment`: - `primary_layer` (e.g., EVIDENCE_CONFLICT, VERIFICATION_NEED) - `verification_trigger` (must be `yes`) - `main_hazard` (the structural danger identified upfront) - `candidate_verification_target` (a rough extraction of what specifically needs checking) If invoked without a clear verification trigger, abort and return to `judgment_hygiene`. ## Verification Target Types Before searching, explicitly classify the object of verification. Search strategies differ by type: - `EVENT`: Did this spec