Polaris: Skill for Claude Code

Verify Spec Closure

Independently certify or reject semantic completion of the exact stable Spec candidate prepared by $verify-spec.

This skill exists to prevent $verify-spec from authorizing its own semantic PASS after it has also selected gates, interpreted failures, repaired the candidate, and assembled proof evidence.

It is not $review-spec:

• $verify-spec-closure asks whether the integrated candidate satisfies the authoritative Spec contract;
• $review-spec remains the later independent Standards / Spec / Architecture adversarial review and convergence layer.

Conditional Claim Discipline

This section is authoritative for any manifest cell whose requirement contains a material condition, trigger, phase, future event, or other predicate controlling when its consequent applies.

Preserve the source condition exactly. Do not silently transform:

when X occurs, Y must happen

into:

build a mechanism now that guarantees Y for every future X

unless the originating Spec itself requires that present mechanism/pre-provisioning.

For every materially conditional cell, record concise proof state:

Condition/trigger: <exact authoritative trigger>
Current trigger state: active | inactive | ambiguous
Trigger evidence: <current evidence>
Deferred routing evidence: <durable destination/owner | None>

Apply these rules:

• active → certify the consequent normally; current behavior must satisfy the cell;
• inactive → current implementation of the consequent is not required unless the source explicitly requires pre-provisioning;
• an inactive cell may be not-applicable for the current candidate only when the originating source establishes the condition and durable decomposition/lifecycle evidence preserves a future destination/owner for the obligation;
• the preferred routing evidence is the parent Spec's current Ticket Coverage Manifest row with a deferred-conditional disposition; equivalent durable authority is acceptable only when it clearly names the future lifecycle/verification destination;
• inactive without durable future routing is unproven, not silently complete, because the obligation would otherwise escape by omission;
• ambiguous trigger state is unproven;
• when the trigger later becomes active, any prior inactive/not-applicable proof is stale by definition and may not be reused;
• the verifier must not invent CI, automation, project policy, a new workflow owner, or another durable mechanism merely to make an inactive conditional cell provable.