Content
71%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A disciplined, dense reference that stays at the decision-rule layer, avoids teaching known concepts, and backstops itself with a concrete evidence-based verification checklist. The two real weaknesses are the absence of any executable examples of the artifacts it governs (JSON-LD blocks, robots.txt, meta tags) and a single-file monolith that contradicts its own declared progressive-disclosure levels.
Suggestions
Split the 'Common Rationalizations', 'Red Flags', and 'Verification' sections into reference files (e.g., references/rationalizations.md, references/verification.md) so the declared progressive_disclosure levels (level2_tokens: 3000) actually exist as files.
Add one minimal copy-paste-ready artifact of each governed output — a sample JSON-LD block, a host-derived robots.txt, and a title/meta-description pair — to lift actionability to fully executable guidance.
Add brief error-recovery guidance to the Verification checklist (what to do when a check fails, e.g., how to reconcile a canonical declaration that redirects) to give the workflow explicit feedback loops.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, assumes Claude's competence (never explains what JSON-LD or robots.txt are), and delivers guidance almost entirely through tight rule tables — 'Normalize before serving. A parameter that does not change the response must not mint a second address'. Minor rhetorical flourishes ('Figures that engines publish and then quietly revise — truncation widths, ranking weights, crawl allowances — are deliberately absent') could be trimmed, so it fits anchor 4 ('efficient; minor instances that could be trimmed') rather than anchor 5, and is clearly above anchor 3 since there is no unnecessary explanation of known concepts. | 4 / 5 |
Actionability | For an instruction-only reference, guidance is concrete and checkable: 'Count h1 elements in the rendered output of each page', 'scan the emitted output for the scaffold's own default strings and fail when one survives', 'give each a stable identifier, and cross-reference by identifier'. It stops short of anchor 5 because there are no copy-paste-ready artifacts — no example JSON-LD block, robots.txt, or meta tag — leaving the emitted form of the output to be inferred. | 4 / 5 |
Workflow Clarity | There is no numbered procedure, but the material is a reference, not a multi-step process; it is organized into a coherent progression (semantics → identity → metadata → structured data → crawl artifacts → delivery) and closes with an explicit Verification checklist demanding observed evidence ('report the counts observed, not the intent', 'quote the address set observed'). That matches anchor 4's 'clear sequence with most checkpoints present'; it falls short of anchor 5 because there are no error-recovery/feedback loops for failed checks. | 4 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are absent) and all ~170 lines of content — including the rationalizations, red-flags, and verification sections — are inline in a single SKILL.md, despite frontmatter declaring 'progressive_disclosure: enabled: true' with a level2_tokens budget. That matches anchor 3 ('some structure... content that should be separate is inline'): section structure is good, but nothing is split out and the declared progressive-disclosure levels have no corresponding files. It is above anchor 2 because the inline content is well-organized rather than a wall of text, and below anchor 4 because no content is actually placed in separate files. | 3 / 5 |
Total | 15 / 20 Passed |