Content
56%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 Adaptyv-specific core is strong - a runnable submit/poll/retrieve example with error handling and clear async-workflow timing guidance - but the body is a merge of two documents, with a generic boilerplate wrapper that duplicates sections, contradicts itself (dependency versions, whether script validation is required), and inflates token cost. The four cited reference/*.md files are missing from the bundle, breaking the progressive-disclosure structure.
Suggestions
Delete the generic template wrapper sections (the first 'When to Use', 'Key Features', 'Dependencies', 'Example Usage', 'Implementation Details', 'Validation Shortcut' plus the boilerplate tail sections: Output Contract, Validation and Safety Rules, Failure Handling, Quick Validation, Deterministic Output Rules, Completion Checklist) - they duplicate the domain content, contradict it (python>=3.9 vs 3.10+, script validation required vs not required), and cost hundreds of tokens without domain value.
Ship the four referenced files (reference/experiments.md, reference/protein_optimization.md, reference/api_reference.md, reference/examples.md) or remove the dangling pointers; as-is, every 'see reference/...' citation dead-ends and the authoritative endpoint/field details the code example defers to are unavailable.
Resolve the workflow contradictions: pick one dependency spec, one validation instruction, and remove the irrelevant date-stamped path in the wrapper's Example Usage ('cd "20260316/scientific-skills/Others/adaptyv"').
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Roughly half the body is generic template boilerplate that adds nothing Claude does not already know: 'Use this skill when the request matches its documented task boundary', 'Keep the output safe, reproducible, and within the documented scope at all times', and entire sections (Completion Checklist, Deterministic Output Rules, Output Contract) of behavioral padding. The Adaptyv-specific core (sections 1-5) is tight, but the padded wrapper sections ('several unnecessary explanations or padded sections') match the score-2 anchor; it is not score 1 because the domain content itself is efficient and free of concept explanations. | 2 / 5 |
Actionability | The core guidance is mostly executable: a complete runnable Python example (env var setup, pip install, bearer auth headers, submit/poll/fetch with raise_for_status and timeouts) plus concrete parameter documentation ('sequences', 'experiment_type', 'webhook_url'). It stops short of score 5 because the example hedges ('Adjust endpoint paths/fields to match reference/api_reference.md', 'endpoint/format may vary') and the authoritative details live in reference files that are not in the bundle. | 4 / 5 |
Workflow Clarity | The experiment workflow is a clear sequence with embedded checkpoints: set credentials -> install -> submit -> poll (or webhook) -> download, with explicit error handling in the code (missing-key RuntimeError, raise_for_status, failure/canceled status break). Not score 5 because validation guidance is inconsistent across the document: the wrapper's 'Validation Shortcut' says to run 'python scripts/validate_skill.py --help' while the later 'Quick Validation' section states 'No local script validation step is required for this skill', and dependencies conflict (3.10+ vs >=3.9). | 4 / 5 |
Progressive Disclosure | The body signals one-level-deep references clearly ('reference/experiments.md', 'reference/api_reference.md', etc.), which is the right pattern, but none of the four referenced files exist in the bundle (only scripts/validate_skill.py is present), so the navigation path is broken and the detail those files should hold (assay types, endpoints, schemas) is only vaguely inline. 'Content that should be separate is inline' from the score-3 anchor applies; it stays at 3 rather than 2 because the signaling and in-body section structure are genuinely clear. | 3 / 5 |
Total | 13 / 20 Passed |