What problem does it solve? Complex, multi-step, or cross-session tasks often fail because the agent acts on unstated assumptions, never tests its beliefs against real evidence, and loses all reasoning state when the session ends. This Skill enforces a disciplined world-model loop so predictions are made before observations, failed predictions actually revise the model, and durable conclusions persist across sessions. ## Core Features & Use Cases - Nine-ring execution protocol: Restore state, build competing models, make falsifiable predictions, gather real observations, evaluate predictions, revise models, and persist results, with CORE and FULL modes gated by task risk. - Persistent state contract: Canonical YAML stores (current model, operators, open loops) plus a runtime append-only ledger and per-session traces, so a new session can mechanically recover every prior conclusion. - Prediction-gated actions: Consequential or high-risk tool calls are blocked unless bound to a specific falsifiable prediction, preventing unexamined mutations. - Use Case: While debugging a recurring cross-session encoding bug, you record two falsifiable predictions before reproducing it; the refuted prediction triggers a structural model revision that is written to the ledger and operators table for future sessions. ## Quick Start Ask the agent to use the world-model-runtime skill to run the full world-model loop on your current high-risk task and persist the resulting state.
