Content
78%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.
An efficiently written, well-structured instruction-only skill: lean prose, a coherent research workflow with feedback loops and explicit stop conditions, and exemplary one-level-deep progressive disclosure into two real reference files. Its main weakness is actionability — much of the guidance is strategic direction without worked examples (sample queries, a filled-in provenance row) that would make it immediately executable.
Suggestions
Add one worked example of an expedition: a starting question, the actual queries tried, and how the gap drove the next query — this would concretize the abstract expedition loop for the actionability dimension.
Include a filled-in example row of the provenance table (e.g., a real-looking source with a Stated vs. Observed distinction and the resulting limit) so the evidence-strength labels are demonstrated, not just defined.
Make the expedition checkpoints explicit validation steps (e.g., "before recording a finding, confirm which label it earns: stated/observed/tested/inferred") to lift workflow clarity from implicit to explicit checkpoints.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~110-line body is dense with directives and free of padding: no explanations of concepts Claude already knows, and every element (provenance table, Stated/Observed/Tested/Inferred taxonomy, stop conditions) earns its place — a clean anchor-5 match, above anchor 4's "minor instances of over-explanation". | 5 / 5 |
Actionability | Concrete elements exist (one retrieval-target reformulation example, a usable provenance record table, a numbered expedition loop), but the guidance is largely strategic: no example search queries, no filled-in example table rows, and no tool-specific invocations, leaving key execution details missing per anchor 3 rather than anchor 4's "mostly executable guidance". | 3 / 5 |
Workflow Clarity | The process is clearly sequenced (establish target → choose surface → expedition loop → provenance → convert to decisions → stop) with a real feedback loop ("choose the next search from the gap") and error recovery ("A correction changes the retrieval target immediately"), but checkpoints like "Inspect enough to support the intended claim" are implicit rather than explicit validation steps, matching anchor 4. | 4 / 5 |
Progressive Disclosure | SKILL.md is a concise overview and both bundle files (references/platform-and-demo-research.md, references/research-integration.md) are real, clearly signaled with conditional read triggers ("For X and visual/Agent research, read..."), and one level deep — the reference files link only to external URLs, never to further bundle files — matching anchor 5. | 5 / 5 |
Total | 17 / 20 Passed |