Classification **Discipline-enforcing.** The rule establishes a hard boundary for coding-task completion claims: the assistant must verify the work before declaring it complete. Rewritten instruction When a bug fix or feature change is presented as complete, you must first run the relevant verification tests; report the task as complete only if those tests pass, and otherwise report that verification failed or remains incomplete. Principles used PrinciplePlacementReasonCommitment“report the task as complete only if…”Creates a clear commitment tied to observable verification rather than an unsupported claim.Authority“you must first run…”Establishes an explicit operational requirement for the coding assistant. What to avoid **Scarcity, Social Proof, Liking, Reciprocity, and fabricated consequences** would backfire here because they add pressure or persuasion where a deterministic verification rule is sufficient. Avoid vague wording such as “try to test,” “when possible,” or “should be complete,” which would create escape routes. Loophole and feasibility check: **One dependency remains:** “relevant verification tests” requires the assistant to determine which tests actually cover the changed behavior. If the necessary tests cannot be run because of missing access, tooling, dependencies, or another external constraint, the assistant must not claim completion and should report the verification limitation. Ethics check **Yes.** The rewrite serves the user's genuine interest by preventing unsupported completion claims without introducing deception, coercion, or fabricated authority. Implementation note Place this in the coding assistant's task-completion or verification rules, and watch for completion reports that explicitly identify the tests run and their pass/fail status.