What problem does it solve? Working in o9's Enterprise Knowledge Graph, a single edit can silently propagate across dependent measures, and a baseline overwrite or plan release to ERP is irreversible without a snapshot. This Skill gives an AI agent the operational judgment to classify every o9 action as read, reversible write, committing write, or destructive, and to gate consequential ones for human approval. ## Core Features & Use Cases - Operation classification matrix: Maps o9 operation families (measure edits, solver runs, scenario commits, version promotion, ERP release) to read/write/destructive gates with reclassification rules based on target version, scope, and snapshot state. - Edge-state and gotcha playbook: Covers 23 real failure modes such as graph propagation, empty spreading bases, shared version-independent measures, concurrent last-write-wins saves, and silent permission drops. - Recovery patterns: Decision tables for undoing bad commits via snapshots, offsetting commits, scenario discards, and ERP-side correction of released orders. - Use Case: Before running a supply solver into the baseline, the agent checks the target version and scope, takes a snapshot, routes a named approver, and re-reads the data to avoid overwriting concurrent planner work. ## Quick Start Ask the agent to run the o9 demand forecast into the baseline version and have it classify the action, take a snapshot, and request approval before executing.
