The body is lean and directive ("Findings are hypotheses, not patches. Verify each against the actual source before applying.") and assumes Claude knows TypeScript, state machines, and review basics, with no padded explanations of concepts Claude already knows. It is efficient rather than the anchor-2 case that includes unnecessary explanation.

Provides a concrete bounded-prompt checklist, named skills-to-load with explicit conditions, a copy-paste output-shape template, and specific commands ("Run package typecheck and tests", "Stage only the files you touched"), giving fully actionable guidance rather than the vague or pseudocode anchors at 1-2.

The "When Editing" section is a 5-step numbered sequence with explicit validation checkpoints ("Add or adjust focused tests first", "Re-read every touched file", "Run package typecheck and tests") and a feedback loop ("Verify each against the actual source before applying"), matching the anchor-3 example with explicit validation steps.

The body is organized into clear sections and references exactly one well-signaled, one-level-deep file ("Read references/type-lifecycle-review.md when the review centers on type protocols..."), which was verified to exist and be self-contained with no nested references. It matches the anchor-3 clear-overview-with-one-level-deep-references example.

Names two review modes ("review and adversarial review") operating on five concrete targets ("concrete diffs, state machines, type shapes, lifecycle boundaries, and confusing abstractions"), matching the anchor for multiple specific concrete actions rather than the anchor-2 example that names only a domain plus a couple actions.

Explicitly answers both what (first sentence on the review scope) and when ("Use when the user asks for..."), with explicit triggers, satisfying the anchor-3 requirement rather than the anchor-2 case where the when is only implied.

Provides direct quoted trigger phrases a user would naturally say ("fresh-context review", "fresh eyes", "an independent reviewer", "adversarial review of a diff", "a state-machine audit"), giving good coverage rather than the partial keyword set of anchor 2.

The explicit negative triggers ("Not for generic helper delegation, executor prompts, ordinary final review, or interactive planning interviews") bound a clear niche and actively reduce conflict risk, matching the anchor-3 distinct-trigger example.