Content
80%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-structured, token-efficient body that delivers exactly the tool-specific knowledge Claude lacks, with a clear workflow and honest output contract. Its weaknesses are the absence of example tool invocations and of explicit validation checkpoints for its batch-mode operations.
Suggestions
Add explicit validation checkpoints for the batch operations, e.g. a step to verify the detected layer count before applying fix_atoms_by_layers, and to confirm selective-dynamics flags were actually dropped after any supercell call before declaring constraints lost.
Include one example invocation per tool (e.g. build_slab with typical slab_thickness/vacuum_thickness values, and fix_atoms_by_layers with a concrete freeze_layers/layer_tol pair) so the guidance is copy-paste ready rather than parameter-level only.
State what to check in the Output Contract deliverables (e.g. how to confirm the chosen termination set is complete for the Miller index) so the final report doubles as a verification step.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence: every line carries tool-specific facts it cannot know ('It emits every termination for the chosen Miller index', 'It drops all selective-dynamics information'), with no padding or background explanation of slab concepts, matching anchor 5. | 5 / 5 |
Actionability | Concrete tool names, parameters (`slab_thickness`, `layer_tol`, `freeze_layers`, `z_ranges`, `centralize`), and hard rules ('freeze_layers must not exceed the detected layer count') give mostly executable guidance, but no example invocations of any tool are provided, so it falls short of the copy-paste-ready anchor 5. | 4 / 5 |
Workflow Clarity | The build → freeze → expand sequence is clearly numbered with real guardrails, but batch operations are explicit ('a whole bulk_dir', 'batch directory mode') and no explicit validation/verification steps exist — checkpoints like the layer-count limit are implicit rules rather than verify steps. Per the guideline capping batch-operation workflows without validation at 3, this matches anchor 3. | 3 / 5 |
Progressive Disclosure | The skill is under 50 lines, needs no external reference files, and is cleanly organized into Overview, Quick Start, Suggested tools, Workflow, Output Contract, and References sections, with a clearly signaled cross-skill handoff ('hand off the post-fix structures to adsorption-site-screening'). Per the under-50-lines guideline this matches anchor 5. | 5 / 5 |
Total | 17 / 20 Passed |