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 strong, highly actionable body with an executable end-to-end example, concrete batching and error-handling constraints, and well-signaled one-level-deep references. Its main weaknesses are duplicated function listings across two sections and the absence of an explicit retry/validation loop for batch operations.
Suggestions
Merge the 'Key Features' function list into the 'API-to-function mapping' section (or vice versa) so each function and its formats are documented once, cutting roughly 15-20 lines of duplication.
Add a short explicit feedback loop for batch operations, e.g., 'If a batch request returns 404, verify each ID with kegg_info and retry the failed IDs individually' to satisfy batch-operation validation expectations.
Move the detailed per-function documentation into references/kegg_reference.md (it duplicates the script's docstrings), keeping only signatures and one-line purposes inline in SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient but includes avoidable duplication: all seven functions ('kegg_info', 'kegg_list', 'kegg_find', 'kegg_get', 'kegg_conv', 'kegg_link', 'kegg_ddi') are listed in 'Key Features' and then re-listed with descriptions in 'Implementation Details', and kegg_get's output formats ('aaseq', 'ntseq', 'mol', 'kcf', 'image', 'kgml', 'json') appear in both sections. This matches the anchor 'mostly efficient but includes some unnecessary explanation or could be tightened' rather than the minor-trim level above. | 3 / 5 |
Actionability | The end-to-end example is fully executable copy-paste Python with real identifiers ('hsa:7157', 'p53', tab-delimited parsing with a guard clause), and the guidance includes concrete batching limits ('up to 10 entries', 'only 1 entry' for image/kgml/json), specific HTTP codes ('400', '404'), and real identifier formats ('map00010', 'cpd:C00002', 'dr:D00001'). Specific examples cover the common cases. | 5 / 5 |
Workflow Clarity | The example demonstrates a clear multi-step sequence (find -> link -> get -> conv) with an inline validation guard ('if not first_line: raise RuntimeError'), and the constraints section defines error handling ('Treat non-200 responses as failures', backoff/retry advice). It falls short of the top anchor because batch operations (chunking lists of <=10 IDs) lack an explicit validate -> fix -> retry feedback loop; it is above anchor 3 because checkpoints and error-code guidance are present, not missing. | 4 / 5 |
Progressive Disclosure | Both referenced bundle files exist (scripts/kegg_api.py and references/kegg_reference.md) and are signaled one level deep with clear purpose ('For detailed endpoint syntax, database lists, and species codes, consult: references/kegg_reference.md'). Minor organization gaps keep it below the top anchor: the inline API-to-function mapping duplicates content available in the script docstrings and reference file, and the function list itself is duplicated across two sections. | 4 / 5 |
Total | 16 / 20 Passed |