skill-world-model

slug	skill-world-model
version	1.0
kind	tool
scope	["chat","simulation"]
summary	世界模型：按 wave 增量构建逻辑自洽的推演画布，覆盖节点本体论、推导语法、上游依据、rationale 与 checkpoint
skillDependencies	[]
capabilityRequirements	[]
assetPolicy	{"references":true,"scripts":false,"templates":false,"assets":false}
name	skill-world-model
description	Logic-rich, wave-driven world modeling for simulation canvases. Use when building, extending, or revising a simulation/world-model canvas that must be semantically incremental, causally traceable, logically self-consistent, and rich enough to cover entities, variables, evidence, hypotheses, inferences, events, risks, scenarios, decisions, actions, outputs, history, and recovery; especially for complex causal questions, market/policy/project risk simulations, scenario planning, node/edge/path generation, upstream-dependent reasoning, checkpoints, reruns, contradiction checks, or explaining why each canvas node exists.

World Model

Purpose

Build a world model as a visible, logically self-consistent reasoning process, not as a fully generated graph hidden behind reveal animation. Generate only the next justified wave of nodes, edges, or paths from confirmed upstream context.

This skill is designed to sit under skill-simulation-base: the simulation base confirms the problem boundary, then this skill constructs the world model wave by wave.

The model must feel full, but not noisy: it should contain enough actors, variables, evidence, mechanisms, risks, paths, decisions, and outputs to explain the problem, while every item remains traceable to an upstream reason.

Operating Rule

Never dump a complete downstream graph in one response. Each response after problem confirmation should produce one coherent wave:

confirmed upstream -> wave analysis -> delta nodes/edges/paths -> rationale -> checkpoint

For current P0 compatibility, emit existing simulation_node, simulation_edge, simulation_path, simulation_summary, and simulation_suggestion blocks. Store wave metadata in data.worldModel on nodes; for edges, put the concise reason in label; for paths, encode traceability in summary unless the runtime adds metadata fields.

Hard constraints:

• One response = one wave delta. Do not output multiple future waves in the same answer.
• Confirmed or locked upstream nodes and edges must never disappear. If an upstream item no longer applies, mark it historical or updated; never omit it as a deletion signal.
• Every node label is the canvas display name. It must be a short noun phrase, not a sentence, statistic, paragraph, or rationale. Put explanatory text in detail; put reasoning in data.worldModel.rationale.
• For entity nodes, label must name the actor/object class, such as 长流程钢企, 废钢电炉钢企, or 铁矿石供应链. Shares, locations, processes, constraints, and uncertainty belong in detail or data.worldModel.uncertainty.
• When emitting or updating simulation_scenario, include scenario.stageState when possible: { current, status, completed, awaitingConfirmation, message, waveId }.