Conciseness

Every section carries library-specific, non-obvious facts (1-based page numbers, percent-0-100 areas, metadata-path resolution, demo-data defaults, controlled-state rules) and nothing explains concepts Claude already knows — PDF, OCR, Svelte, or JSON Schema basics are never introduced. This is the anchor-5 'lean and efficient; every token earns its place' case, not anchor 4, since no over-explanation instances stand out to trim.

Actionability

Four complete, copy-paste-ready Svelte 5 examples with real imports, props, and bind directives cover the common cases for each component, and concrete helper calls (serializeSchema, getOcrBlocks, blockToArea, createInitialSplits) plus exact import paths are given throughout. This matches anchor 5's 'fully executable; copy-paste ready code; specific examples cover the common cases' better than anchor 4, which would allow gaps in the examples.

Workflow Clarity

The flagship flow is a clear 5-step sequence (map output → supply locations → bind activeFieldKey → bind values/submit → handle callbacks) with useful guardrails ('keep the array reference stable', inline location wins over metadata), matching anchor 4's 'clear sequence with most checkpoints present'. It is not anchor 5 because there are no explicit validation/verification checkpoints (e.g., how to confirm citations render before wiring submission), and step 3 points to an example file that is not in the bundle.

Progressive Disclosure

The body is a well-signaled overview with a References section whose four reference files all exist in the bundle and are one level deep — close to anchor 5. However, `examples/extraction-review.svelte` and `examples/document-intake.svelte` are cited twice in the body ('Full mock-data screen: examples/extraction-review.svelte', 'See examples/document-intake.svelte') but are absent from the bundle's file listing, a minor organization gap that fits anchor 4's 'minor organization gaps' better than anchor 5's 'easy navigation'.

Specificity

The description lists multiple concrete actions — 'build an extraction schema editor or JSON schema builder', 'split a multi-document PDF into sections', 'show parsed OCR layout blocks', 'show extracted values with citations or bounding boxes on a PDF', 'highlight the source region of an extracted field', 'compare actual vs expected extraction output' — giving comprehensive coverage of the skill's capabilities with no material gaps, matching the top anchor rather than anchor 4.

Completeness

It explicitly answers both questions: 'This skill should be used when the user asks to [long list of concrete trigger phrases]' is an explicit when-clause, and the what is stated as building a 'document-AI / extraction-review screen with the extend-ui-svelte Svelte 5 library' with named components. Voice is third person ('the user asks'), so no person-penalty applies; this is a clear anchor-5 match, not anchor 4 where the when-clause would be less explicit.

Trigger Term Quality

Natural trigger phrases are comprehensive and paired with synonyms users would actually say: 'schema editor'/'JSON schema builder', 'sections'/'document splits', 'human review'/'document review UI', 'citations'/'bounding boxes', plus 'OCR layout blocks' and 'actual vs expected extraction output'. This matches anchor 5 (comprehensive natural terms including synonyms) rather than anchor 4, which requires missing terms.

Distinctiveness Conflict Risk

The niche (document-AI / extraction-review UI with extend-ui-svelte, component names listed) is clear and mostly distinct, but the phrases 'build a … JSON schema builder' and 'build a human review … UI' are broad enough to attract generic non-library requests — minor overlap risk with closely related skills, matching anchor 4 rather than anchor 5's 'minimal conflict risk'.