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.
The body is a lean, well-structured instruction-only router skill with a clear sequenced workflow and pre/post checkpoints, and appropriately needs no bundle files. Its main weakness is actionability: the routing policy is stated abstractly with no worked request-to-tool examples and no tie-breaking rule for overlapping tools.
Suggestions
Add 2-3 worked routing examples mapping representative requests to specific tools (e.g., 'Edit a shape on a slide → design_update_shapes; find an integration → mcp_search') to make the guidance executable rather than policy-like.
Add a tie-breaking rule for overlapping tools (e.g., when browser_use vs mcp_search, or office_inspect vs ppt_agent, both seem applicable) so step 2 ('Choose the most specific advertised tool') has a deterministic procedure behind it.
Tighten the residual slack: drop the Purpose line that duplicates the frontmatter description and the 'Delivery' section's general reporting advice, both of which restate what Claude already does by default.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean, assumes Claude's competence, and explains no known concepts; the tool table is compact and directive. Minor trim candidates keep it below 5: the Purpose line duplicates the description verbatim, and the 'Delivery' section states general agent behavior ('Lead with the outcome, name the evidence') that Claude already knows. | 4 / 5 |
Actionability | There is some concrete guidance — a six-row tool-to-use table and a concrete output-classification list ('prose, file, code, browser action, design, chart, diagram, presentation, media, schedule, or integration') — but it is incomplete: no worked examples of a request routing to a specific tool, and no tie-breaking rule for overlapping tools (e.g., browser_use vs mcp_search). Not 4 because key execution details a router skill needs are missing. | 3 / 5 |
Workflow Clarity | The five-step sequence is clear with checkpoints present: 'Inspect current state before mutation' (step 3) and 'Verify using the closest observable output' (step 5), plus 'Report unresolved failures without hiding them'. Not 5 because the verification step is abstract with no fix-and-retry feedback loop and no explicit error-recovery path. | 4 / 5 |
Progressive Disclosure | The skill is under 50 lines, single-purpose, has no bundle files (no references/, scripts/, or assets/ exist), and its sections (Purpose, Tool routing, Workflow, Completion gates, Boundaries, Delivery) are well-organized with no content that belongs in a separate file. The simple-skill exception applies: well-organized sections alone merit 5. | 5 / 5 |
Total | 16 / 20 Passed |