Model-check the plan before any tool runs.

arXiv 2609.18674 extends CaMeL with a static verification layer. CaMeLoT translates a generated plan into a finite-state transition system labeled with tool calls, provenance and taint, then checks it against CTL policies with nuXmv before execution starts. Unsafe plans get rejected with zero LLM calls, zero tool calls, no sandbox teardown and no state to unwind, and a failed check returns a counterexample the agent can use to repair its plan. Policies come from AgentDojo, SOC workflows and prompt-extraction experiments. Verification before the first side effect is cheaper than any rollback.