Content
92%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 well-engineered instruction skill: an explicit 7-step workflow with a real review-checkpoint step, component-level actionability throughout, and exemplary progressive disclosure (purpose-labeled references, one level deep, with a conditional-load gate for the Genie material). The only weakness is token redundancy — the body's routing section and Genie gate repeat content already carried by the always-loaded description and by the section's own opener.
Suggestions
Trim the 'When to use / when NOT' section to only the deltas from the frontmatter description (e.g., keep the concrete non-trigger examples like '"add a form", "deploy this"' but drop the restated Lakeview-routing paragraph, which the description already states verbatim).
State the Genie gate once: the section opener ('Gate: this section applies only if...') and the following sentence ('When it does apply, implement ALL five') both establish applicability — merge into a single gating sentence.
Merge workflow step 6 ('Cover the states') with the 'Required states & data realism' section, or reduce one to a pointer, since loading/empty/error/partial is enumerated in both places.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and prescriptive with zero concept-explanation padding and delegates domain depth to references, but the 'When to use / when NOT' section restates the frontmatter description's routing paragraph nearly wholesale ('A plain "create a dashboard" / "build a dashboard" request (no app / AppKit / React / custom-code signal) means a managed AI/BI (Lakeview) dashboard'), and the Genie gate is stated twice within the same section ('Gate: this section applies only if...' followed by 'For ANY Genie/chat surface, ship all five'). These are minor redundant tokens that could be trimmed — anchor 4 ('efficient; minor instances... that could be trimmed'), not 5 ('every token earns its place'). | 4 / 5 |
Actionability | Instruction-only skill whose guidance is fully actionable: names exact primitives ('Skeleton', 'Empty', 'Badge', 'Card', 'Alert'), hooks and data binding ('useAnalyticsQuery'/'queryKey' + 'sql.*' params, 'useGenieChat().status'), tokens ('colorPalette + semantic tokens, never hardcoded hex'), concrete API surface ('/api/whoami' route, 'x-forwarded-email'/'x-forwarded-user' headers, 'attachments[].query', 'user_api_scopes: [dashboards.genie]', 'genie()' space config), plus copy-paste-ready output templates and pointers to exact snippets ('copy the exact snippets from the reference'). Per the code_vs_instruction scoring note, absence of inline code is not penalized when guidance is this concrete. | 5 / 5 |
Workflow Clarity | Seven clearly sequenced steps (Frame → Genre → Compose → Apply notation → Bind to components → Cover the states → Review) with an explicit validation checkpoint — step 7 'run the checklists in both reference files', and both checklists verifiably exist (dashboard-patterns.md '## Review checklist', ibcs-notation.md '## Review checklist' + '## SUCCESS checklist'). The critique format requires 'the concrete fix (which component/token/state to change)' per finding, closing the design→review→revise loop. Not 4: checkpoints are present, not 'mostly' present; the destructive/batch cap does not apply to a design/review skill. | 5 / 5 |
Progressive Disclosure | Clean overview with four one-level-deep references, each purpose-labeled at the point of link ('**Composition** — what to show... → references/dashboard-patterns.md'; '**Notation**... → references/ibcs-notation.md'; '**Implementation**... → references/appkit-cheatsheet.md'), all verified present in the bundle. Conditional loading is handled exceptionally: 'skip this section and references/genie-ai-trust.md entirely' gates an entire reference file. Reference files cross-link only to siblings or the declared parent skill — no 2+-level nesting. Anchor 5 ('clear overview with well-signaled one-level-deep references') matches. | 5 / 5 |
Total | 19 / 20 Passed |