What problem does it solve? FourKites is a real-time transportation visibility platform where almost every operation is a read, so the real risks are not posting money or stock but leaking sensitive shipment, customer, and driver location data, acting on unconfirmed predictive ETAs or self-reported milestones, and making committing writes like dock appointments or cross-system status pushes. This Skill gives an AI agent the operator judgment to classify every action by data sensitivity and egress, and to pause consequential ones for human approval. ## Core Features & Use Cases - Action classification matrix: Categorizes FourKites operations into read, reversible local write, committing write, and destructive/high-sensitivity classes, each with a defined gate level from pass-through to hard gate with named owner. - Egress and party-scope control: Treats any output crossing a tenant, party, or audience boundary (including AI assistant replies) as a controlled disclosure, with a step-by-step egress decision procedure. - Mode-specific reference depth: Covers OTR, LTL, Dynamic Ocean (LFD, demurrage/detention), air, parcel, and rail milestones, EDI 214 codes, Tracking Score interpretation, and Appointment Manager / Dynamic Yard committing writes. - Use Case: Before pushing a predictive ETA into a TMS that auto-re-tenders loads, the agent recognizes the push as destructive-by-proxy, corroborates the ETA against a fresh GPS ping, and requires named human approval. ## Quick Start Ask the agent to check whether a specific FourKites load's predictive ETA is reliable enough to promise a customer delivery, and have it classify the action before responding.
