Content
46%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 core PubMed-specific material (query construction, input collection, retrieval, topic-generation rules, references) is solid and well-linked to real bundle files, but it is buried under generic template sections that add no task-specific value, and the quick-validation instructions contradict the actual script interface. The skill would be markedly stronger at roughly half its current length with the validation step rewritten against the script's real CLI-less configuration flow.
Suggestions
Cut the generic boilerplate sections (When Not to Use, Required Inputs, Recommended Workflow, Deterministic Output Rules, Output Contract, Validation and Safety Rules, Failure Handling, Completion Checklist, Scope Reminder) and keep only the PubMed-specific workflow: configure CONFIG in scripts/run_topic_recommendation.py, run it, read output/topic_recommendations.json.
Fix the Quick Validation section: the script has no argument parsing, so "python scripts/run_topic_recommendation.py --help" will not print usage — replace it with a real check (e.g., confirming CONFIG['direction'] and keywords are filled before running) and correct the expected output filename to match the script's actual output/topic_recommendations.json.
Replace the hedged "names may vary by implementation" / "implementation-dependent" language with the script's actual configuration keys (direction, keywords, exclude_keywords, start_date/end_date, article_types, recommendation_count) so the setup steps are copy-paste actionable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Roughly half the body is generic template filler not specific to this task: "Do not use this skill when the required source data, identifiers, files, or credentials are missing", "Use the same section order for every supported request", and a "Scope Reminder" that repeats the description verbatim. Not 3 because the padding spans multiple full sections (Required Inputs, Deterministic Output Rules, Output Contract, Validation and Safety Rules, Failure Handling, Completion Checklist), and "Key Features" duplicates "Implementation Details". | 2 / 5 |
Actionability | There is a real executable command ("python scripts/run_topic_recommendation.py") and a real configuration area in the script, but guidance is hedged ("names may vary by implementation", "A JSON file or JSON printed to stdout (implementation-dependent)") and partly inaccurate: "python scripts/run_topic_recommendation.py --help" fails because the script takes no CLI arguments, and the claimed result file "pubmed_topic_recommend_result.md" contradicts the script's actual "output/topic_recommendations.json". Not 4 because the inaccurate quick-validation path and implementation-dependent hedges leave a user unsure of exact behavior. | 3 / 5 |
Workflow Clarity | A configure-then-run-then-output sequence exists, and there is a genuine retrieval feedback loop ("If results are too few: broaden date range, relax field restrictions, or add synonyms"). However the "Recommended Workflow" is boilerplate ("Select the documented execution path and prefer the simplest supported command") rather than task-specific checkpoints, and the validation step relies on the nonexistent --help flag. Not 4 because the primary validation checkpoint is misleading rather than just minor-gapped. | 3 / 5 |
Progressive Disclosure | Clear section structure, and detail is appropriately split into one-level-deep, real, clearly signaled references: "API and query syntax reference: references/pubmed_api.md" and scripts/run_topic_recommendation.py, both of which exist in the bundle. Not 5 because some inlined boilerplate (Deterministic Output Rules, Completion Checklist) adds noise, and the script's configuration options are summarized in the body rather than pointed at cleanly. | 4 / 5 |
Total | 12 / 20 Passed |