Content
70%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.
A well-sequenced, information-dense workflow with excellent gating logic, conflict-resolution loops, and error visibility, but held back by malformed JSON in several curl payloads that prevents copy-paste execution and by a fully monolithic structure with no offloaded reference files. The duplicated api_calls schema and Tips restatements add minor token overhead.
Suggestions
Fix the malformed JSON payloads in Apollo people/match, both Sixtyfour calls, and Brand.dev ai/products by wrapping parameters in a "query":{...} object, matching the working Hunter examples (e.g., -d '{"api":"apollo","path":"/api/v1/people/match","query":{"email":"{email}","reveal_personal_emails":true}}').
Move the full output-format JSON schema, cost-tracking table, and per-provider field-extraction lists into a references/ file (e.g., references/output-format.md), keeping SKILL.md as a concise overview with well-signaled one-level-deep links.
Deduplicate the meta.api_calls structure (define it once in Error Visibility and reference it from Output Format) and trim Tips entries that restate gating already specified per phase.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with operational specifics Claude cannot know (endpoints, field paths, merge rules, per-call costs) with no padding or basic-concept explanations. It is not 5 because the meta.api_calls schema is specified twice (once inside the Output Format example and again in Error Visibility) and the Tips section restates gating logic already given per phase. | 4 / 5 |
Actionability | Most curl commands are complete and executable (Hunter combined/verifier, Apollo org enrich, Tomba, GitHub), but the Apollo people/match, both Sixtyfour, and Brand.dev ai/products payloads are malformed JSON — the -d string closes early and dangles "email": "{email}" without the "query":{...} wrapper used in the working examples. It is not 4 because several commands are not copy-paste runnable, and not 2 because the majority of guidance is concrete with fully specified extraction fields. | 3 / 5 |
Workflow Clarity | Phases 0-5 are explicitly sequenced with precise gating conditions (free-email check, funding-missing, person-data conflict, person-not-found, funded+B2B+>50 employees), a genuine conflict-resolution feedback loop (Tomba tie-breaker with keep-both-and-flag on three-way disagreement), and an explicit error taxonomy (success/partial/error/skipped) with the rule "Never silently skip failures". It is not 4 because validation checkpoints and error-recovery behavior are explicit at every phase boundary, not merely present for most steps. | 5 / 5 |
Progressive Disclosure | The ~420-line body has clear section headers and a coherent flow, but no bundle files exist and everything is inlined, including content that clearly belongs in separate files (the ~60-line output JSON schema, the cost-tracking table, per-provider field maps). It is not 4 because substantial reference material is inlined in SKILL.md rather than split out, and not 2 because the inlined content is well structured, not a minimal-structure wall. | 3 / 5 |
Total | 15 / 20 Passed |