Structure Review

Overview

adversarial-review asks whether the artifact is wrong. edge-case-hunter asks what inputs it ignores. This skill asks a third, orthogonal question: does the artifact cohere — is it organised, complete, and internally consistent as a document, regardless of whether any individual claim is correct? A plan can be factually right in every sentence and still fail here, because two sections contradict each other or a promised section is missing.

What It Checks

1. Completeness against the artifact's own contract. If it is a plan, does every declared component have implementation steps and a validation entry? If it is a PRD, does every acceptance criterion trace to a requirement? The test is closure: no dangling reference, no promised-but-absent section, no requirement with no owner.
2. Internal consistency. Grep the artifact against itself. Does §3 assume what §7 rules out? Does the summary count differ from the enumerated list? Does a "MUST" in one place become a "should" in another? Contradictions between sections are the highest-yield structural defect.
3. Traceability. Can a reader follow each conclusion back to its premise and each requirement forward to its verification? A decision with no recorded reason is a broken chain.
4. Runnable step order. Steps that reference outputs produced only by later steps, or tools installed only afterwards, are un-runnable as written — the document lies about its own execution order.

When It Loads

• /dr-plan — before a plan is frozen, verify it holds together as a document.
• /dr-qa Layer 3 — structural/editorial pass over the produced artifacts.