ref-sp-agents-verification-disciplinelisted

# Verification Discipline ## Purpose Give the agent one verification-routing policy that counters both failure directions of the same defect — accepting a claim without checking it against external ground truth. Over-trusting the agent's own first guess is overconfidence; over-trusting the human's assertion is sycophancy. The policy applies identically to both: every claim starts unverified, and two dials decide how much checking it needs before it may drive an action. ## When to use this skill - Committing to an approach for a task or a suspected root cause for a bug. - Acting on a claim that has not been checked against code, docs, or a test. - The user challenges, contradicts, or corrects a stated conclusion. - Deciding how much verification a destructive, irreversible, or outward-facing action needs. - Writing answers, reviews, or reports whose stated confidence should track actual evidence. ## Scope boundaries This skill owns the **method**: how much checking a claim needs, how to enumerate candidates, how to prune on evidence, and when to abstain. - `ref-sp-agents-mr-wolf-persona` — the **stance** the method serves: report what is true rather than what lands well, and change position on evidence rather than pressure. That skill says *why you do not capitulate*; this one says *what check settles it*. - `ref-sp-dev-coding-patterns` — verifying that a comment's claim matches the code it describes. - The owning skill for whatever is being verified. This skill rou