Content
42%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 raw API dump: duplicated parameter lists, several malformed curl examples, and a truncated section, with the async-research workflow only implied. It needs copy-paste-ready examples, deduplicated parameters moved to a reference file, and an explicit task-polling loop.
Suggestions
Fix the broken curl blocks in Search, Find similar, and Get contents so the JSON body is inside the -d string (e.g. -d '{"api":"exa","path":"/search","body":{"query":"...","numResults":10,"contents":{"text":true}}}'), and use consistent parameter names (numResults vs num_results, urls vs ids).
Move the full per-endpoint parameter reference into a separate references/ file (e.g. references/api.md) and keep only a working example plus 2-3 key parameters per endpoint in SKILL.md; this also removes the verbatim duplication between Search and Find similar.
Complete the truncated Answer description and add an explicit research-task workflow: create task → save researchId → poll GET /research/v1/{researchId} until status is finished → read results, including what to do on failure or timeout.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The "Find similar links" section repeats the entire Search filter parameter list verbatim (includeDomains, excludeDomains, crawl/published dates, includeText, excludeText, context, moderation — ~14 duplicated lines), and parameter descriptions repeat the same boilerplate text three times over. Not anchor 3: this is systematic duplication and padding, not occasional looseness. | 2 / 5 |
Actionability | Setup and four curl examples are concrete, but the Search, Find similar, and Get contents examples are syntactically broken — the JSON body ("query": ..., "num_results": ...) sits outside the -d quoted string, leaving a stray '}' — and the Answer description cuts off mid-sentence ("A direct answer for specific queries. (i.e."). Not anchor 4: a third of the code blocks are not executable as written. | 3 / 5 |
Workflow Clarity | A rough sequence exists for research tasks ("Use the unique researchId returned from POST /research/v1 to poll until the task is finished") but steps are scattered across endpoint sections with no explicit checkpoints, no polling-loop guidance, and no error recovery. Not anchor 4: checkpoints for the async create→poll→fetch flow are implicit at best. | 3 / 5 |
Progressive Disclosure | Section headers (Capabilities, Usage, per-endpoint subsections) provide navigation, but ~200 lines of full API parameter reference are inlined in SKILL.md with no separate reference files and no pointers to move detail off the main page. Not anchor 2 because real structure exists; not anchor 4 because the bulk parameter documentation clearly belongs in a reference file. | 3 / 5 |
Total | 11 / 20 Passed |