Content
50%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 a broad, well-organized API catalog with concrete commands and clear per-task routing ('Best for', Tips), but it is not copy-paste reliable: multiple curl examples have JSON bodies placed outside the -d argument, and the closing 'Discover More' section is garbled. Combined with ~40-fold duplication of the auth boilerplate, zero error-handling guidance for async polls and batch jobs, and no use of reference files for the five API catalogs, the skill is serviceable but needs tightening and verification steps.
Suggestions
Fix the malformed curl examples: in the smartscraper, crawl, riveter /v1/run, and batch blocks, the JSON body appears outside the -d '...' argument (the quote closes at "path":"/v1/..." and body keys dangle unquoted); rewrite each as a single complete -d '{...}' payload, and clean up the 'Discover More' section where stray text ('api show olostep') and a stray backtick break the commands.
State the shared curl invocation once (base URL, Bearer auth header, api/path/body structure) and show only the per-endpoint JSON payloads in the examples — this removes the four-line header block repeated ~40 times and roughly halves the file's token cost.
Add validation and error-recovery guidance for the async and batch workflows: how to interpret poll responses (pending vs. completed vs. failed), what to do on error, and a checkpoint before retrieving batch/crawl results.
Move the full endpoint catalogs for each API (Scrapegraph, Olostep, Riveter, Brand.dev, Notte) into per-API reference files under references/, keeping SKILL.md as an overview with 'Best for' routing and one or two key examples per API.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The identical four-line curl auth-header block (URL, Authorization, Content-Type, api/path prefix) is repeated in nearly every one of ~40 examples, inflating the file well past what the content earns. It avoids explaining concepts Claude already knows, but the boilerplate duplication means it 'could be tightened' — factoring the shared invocation into one stated pattern would cut the file roughly in half. | 3 / 5 |
Actionability | Commands are concrete curl calls, but a significant number are malformed and not executable as written: several blocks (e.g., the smartscraper, crawl, and riveter /v1/run examples) close the -d quote before the JSON body, leaving body keys like "website_url" outside the request; the 'Discover More' section appends stray text ('api show olostep') after the -d argument and contains a stray backtick. This is 'concrete guidance but incomplete; missing key details' rather than mostly-executable with minor gaps. | 3 / 5 |
Workflow Clarity | Multi-step flows are clearly labeled (Notte 'Step 1–5', Olostep crawl start/status/pages/retrieve), giving a real sequence. However, validation is absent throughout: no guidance on checking poll responses for errors or completion, no error-recovery loop, and the batch-scraping operations (Olostep batches, async crawls) run with no verification steps — batch operations without validation cap this dimension at 3 per the rubric. | 3 / 5 |
Progressive Disclosure | The file has good section structure (numbered per-API sections with 'Best for' lines and a Tips section), but there are no bundle files at all: full endpoint catalogs for five distinct APIs (~440 lines) are inlined in SKILL.md, where the rubric expects that bulk to live one level deep in reference files. This fits 'some structure but content that should be separate is inline'. | 3 / 5 |
Total | 12 / 20 Passed |