Content
75%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-built, domain-aware instruction skill: the escalation workflow with concrete supercell values, the guardrails separating visual from numerical evidence, and the explicit Output Contract are all strong. The weaknesses are structural rather than substantive — Quick Start duplicates the Workflow, 'Suggested tools' duplicates frontmatter metadata, and the 'References' header wraps a closing remark instead of actual references.
Suggestions
Merge Quick Start into the Workflow (or reduce Quick Start to a one-line pointer) to remove the duplicated render-first/escalate guidance.
Drop the 'Suggested tools' section or the frontmatter catmaster-suggested-tools metadata — one of the two is redundant.
Rename or repurpose 'References' — either point to real reference files (e.g., a supercell/site-family guide) or fold its caveat into 'Method-critical defaults'.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean, assumes domain competence, and explains nothing Claude already knows — every section is operational guidance. Not 5: there is real duplication (Quick Start items 1-2 restate Workflow items 1-3, and 'Suggested tools' repeats the frontmatter tool list), and the 'References' section is a single closing sentence that earns no tokens under that header. Not 3: the padding is minor and local, not 'some unnecessary explanation' spread throughout. | 4 / 5 |
Actionability | Guidance is concrete for an instruction-only skill: 'render a default four-view panel first', 'Typical slab retries are (2,2,1) or (3,3,1)', 'Use analyze_images only with a narrow question', 'keep the rendering preset and visual context aligned across all images', 'return the main panel image path'. Not 5: a few directions stay high-level — no example invocation or parameter snippet for render_structure_views/analyze_images, and 'hand back to numerical geometry tools or file-based metadata' names no concrete tool or step. Not 3: nothing is pseudocode or abstract; the specific supercell values and output contract are directly executable. | 4 / 5 |
Workflow Clarity | The Workflow section is a clear escalation sequence (default render -> supercell retry (2,2,1)/(3,3,1) -> narrow image analysis -> aligned comparison -> hand back to numerical tools) with explicit verification checkpoints: 'Confirm critical geometric claims with numerical tools when available' and 'verify it with file-based structures or numeric geometry analysis before changing the workflow'. Not 5: the checkpoints stop short of a true feedback loop — there is no instruction for what to do when the numeric check contradicts the visual conclusion, and the Output Contract is not tied back into the workflow steps. Not 3: checkpoints are explicit, not merely implicit; this is not a destructive/batch operation so the cap does not apply. | 4 / 5 |
Progressive Disclosure | No bundle files exist, and the ~40-line body is organized into well-labeled sections (Quick Start, Suggested tools, Workflow, Method-critical defaults, Output Contract) — under the simple-skill guidance this structure is appropriate. Not 5: the 'References' section is misused — it contains a closing caveat instead of referencing any file, and 'Suggested tools' duplicates frontmatter metadata rather than adding anything, both minor organization gaps. Not 3: there is no content that should be split out to separate files; everything inline belongs inline. | 4 / 5 |
Total | 16 / 20 Passed |