Content
50%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A detailed, well-sequenced operational skill whose main weakness is execution correctness: every multi-field curl payload has the same JSON syntax error. Beyond that, a large batch workflow lacks explicit validation checkpoints and the single monolithic file foregoes progressive disclosure.
Suggestions
Fix the malformed `-d` JSON in every curl example: the proxy wrapper object must not close before the request body fields — merge `api`/`path` with `website_url`/`user_prompt`/`body` etc. inside one JSON object (e.g. `-d '{"api":"scrapegraph","path":"/v1/smartscraper","website_url":"...","user_prompt":"..."}'`).
Add an explicit validation checkpoint in the batch loop — e.g. after each company's research, verify the row wrote successfully and re-trigger any failed scrape/Apollo call before moving to the next batch — to satisfy the batch-operations feedback-loop requirement.
Split the large body into one-level-deep reference files (e.g. references/api-reference.md for endpoint payloads, references/formatting-rules.md for column layout and link formatting, references/scoring-rubric.md for fit/overall/rank guidance) and keep SKILL.md as a concise overview that links to them.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Almost all content is API-specific operational detail Claude does not already know (endpoints, field-name variants, column layouts), so it avoids explaining known concepts; however the 'live-fill demo effect' point is restated three times and several emphatic 'do NOT'/'do not skip' directives are redundant, so it could be tightened. | 2 / 3 |
Actionability | Commands are highly concrete with exact endpoints, expected response shapes, and column mappings, but every multi-field curl `-d` payload is malformed JSON (the wrapper object closes with `}` before `website_url`/`user_prompt`, and those fields sit outside the quoted string), so the examples are not copy-paste executable as written. | 2 / 3 |
Workflow Clarity | Steps 1–5 are clearly sequenced with a parallelization strategy and per-failure fallbacks, but this batch operation over ~22 companies has no explicit validate-then-proceed checkpoint, which caps workflow clarity at 2 per the batch-operations rule. | 2 / 3 |
Progressive Disclosure | The ~510-line body is well-sectioned but monolithic with no bundle files; the API reference, formatting rules, fit/scoring rubrics, and cost table are inline content that would navigate better split into one-level-deep reference files. | 2 / 3 |
Total | 8 / 12 Passed |