@schchit

determinability-checker gates skill execution by proving whether a judgment question is answerable from current evidence. it returns DETERMINED or NOT_DETERMINED with counterexamples and missing-evidence guidance, implementing finite model checking from causal observation theory.

coe-consensus orchestrates multi-agent world state agreement via pluggable consensus policies (simple majority, weighted trust, byzantine fault tolerance). produces verifiable shared world state records from heterogeneous sensor and judgment events.

jep-guard-audit provides a three-layer audit chain mapper translating developer-friendly events into strict jep-04 compliance records, with export adapters for eu ai act, california sb 1047, and colorado sb 205 standards.

claimback-radar extracts billing data from emails and bills via openai api to identify refund opportunities and subscription risks, but lacks concrete procedural steps and trigger definitions.

jep-guard provides pre-execution interception and audit logging for high-risk operations across skill chains, with reputation tracking and jep-standard event export. concrete implementation details are sparse.

JEP Primitive Skills — Atomic Reference Implementations of Judge, Delegate, Terminate, Verify for Agent Collaboration Grammar