What problem does it solve? Autonomous research pipelines can over-escalate routine questions to an external Oracle advisor, wasting consultations and creating inconsistent decision records. This Skill enforces a strict eligibility gate so that only genuinely unresolved, high-impact scientific decisions trigger an Oracle/ChatGPT Pro consultation. ## Core Features & Use Cases - Five-Condition Eligibility Gate: Returns consult_oracle only when a concrete pending decision has two or three defensible alternatives, material consequences, exhausted local evidence, and an actionable recommendation path; otherwise returns continue_local. - Consult-Once Deduplication: Assigns semantic decision ids (e.g., mapping:mechanism-family:v1) and reuses recorded advice when no new evidence changes the alternatives. - Bounded Evidence Packets: Prepares a structured packet with alternatives, baseline and literature evidence, unresolved uncertainty, and one to eight workspace-relative evidence files before invoking the oracle_governor runtime tool. - Use Case: During an autonomous research run, a phase skill cannot decide between two literature-supported mechanism families that determine the central hypothesis; this Skill validates eligibility, submits one bounded Oracle consultation, and records the recommendation in research/CONTRIBUTION_MAP.md. ## Quick Start Ask the agent to evaluate whether the current unresolved research decision qualifies for an Oracle consultation and, if eligible, submit it through the oracle_governor tool.
