validate-inferencelisted

# Validate Inference (before acting on it) ## Overview The failure mode this skill interrupts: an assistant infers a premise (most commonly "nothing changed since the prior X, so skip / proceed"), acts on the inference, and the action propagates a wrong premise into downstream work. By the time a later check catches it, the cascade has touched multiple artefacts, and the rework costs orders of magnitude more than the single tool call that would have validated the premise at its source. An inferred premise is a claim about state that the next action depends on and that has not been directly observed in the current turn. The trigger surface is textual and catchable in your own draft: clauses of the form "since / because / given / per / based on X, [action]" where X has not been verified this turn. The canonical rule ([`validate-inference-before-action`](../../governance/validate-inference-before-action.md)) states the discipline; this skill is its workflow wrapper for the moment the trigger fires. ## When to Use - A draft or plan contains "since / because / given / per / based on / following / according to" followed by a state claim not observed this turn. - About to SKIP a step (a subagent, a test, a gate, a review) on the grounds that a prior run covered it or that nothing relevant changed. - About to declare a fix complete because the named occurrence is fixed (the parallel-occurrence inference). - About to reuse a prior approval, a prior file read, or a prior too