Content
88%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 body is an unusually strong instruction-only skill: exact tool and field names, an explicit status-check feedback loop, and honest, measurement-backed limits (latency percentiles, enum truncation, retailer field-coverage variance) that surface real failure modes. The only gaps are minor: inline date-stamped measurements and a length that slightly exceeds what should live in one file.
Suggestions
Move the measured limits (latency percentiles, the 2026-08-27 marketplaces enum-truncation measurement, field-coverage counts) into a short 'Measured behavior' or reference file with a note of when they were captured, so the main body stays date-stable.
Split the field-quirk reference ('The four differences that bite hardest' plus the additionalProperties details) into a one-level-deep reference file linked from the body, bringing SKILL.md under the ~50-line simple-skill threshold.
Add one fully assembled example input object (combining keyword, marketplaces, maxProductResults, and additionalProperties) so the first call can be made verbatim without composing from the input table.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence — no explanation of what MCP or product data is, and every section reports tested findings (e.g. "one Amazon product measured about 88 KB across 142 fields"). One point off because time-sensitive measurements carry inline date stamps ("Measured on 2026-08-27: of 249 values, 122 fit and 127 were dropped"), which the guidelines penalize outside a staleness/deprecated section; it is not a 3 because there is no padded explanation anywhere. | 4 / 5 |
Actionability | Guidance is fully concrete: exact tool name with its gotcha ("apify--e-commerce-scraping-tool (two hyphens, not a slash)"), exact inputs ("detailsUrls: [{\"url\": \"...\"}]", "additionalProperties: true"), exact flat key access ("item[\"offers.price\"]"), and named fields (inStock, stars, offers.priceCurrency). For an instruction-only skill this is copy-paste-ready guidance covering the common cases, matching the level-5 anchor. | 5 / 5 |
Workflow Clarity | The three-step sequence has an explicit validation checkpoint ("Check status. If it is not SUCCEEDED, call get-actor-run with the runId and a waitSecs until it is") with a feedback loop, plus named failure modes ("Stopping after step 1...", "Skipping step 2 fetches an empty dataset") and recovery paths (marketplaces validation error → use detailsUrls). This matches the level-5 anchor; the operations are read-only lookups so no destructive-operation cap applies. | 5 / 5 |
Progressive Disclosure | The single-file skill is well organized into clearly signaled sections (Setup, input selection, call steps, field projection, honesty rules, limits), and there are no broken references since no bundle files exist. It falls at level 4 rather than 5 because the body runs ~97 lines of dense reference material (field quirks, scope-and-limits measurements) — above the under-50-line simple-skill threshold — that could be split into a one-level-deep reference file. | 4 / 5 |
Total | 18 / 20 Passed |