Content
63%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 dense, highly actionable CLI reference with realistic commands, sensible async polling patterns, and a genuine error-recovery section — its main strengths are actionability and workflow clarity. Its weaknesses are structural: a single monolithic file with no reference files for per-subcommand detail, plus repeated when-to-use and pitfalls content that inflates token cost without adding guidance.
Suggestions
Split the FindAll, Monitor, and Enrichment sections into references/ files (e.g. references/findall.md) and keep one-line pointers in SKILL.md, moving the skill toward one-level-deep progressive disclosure.
Consolidate the tripled when-to-use guidance ("Important expectations", "When to use it", "Pitfalls") and the Core-rule-set/Pitfalls overlap into a single section to reduce token cost.
Provide a concrete `monitor create` example (cadence, delivery target, URL) instead of deferring to `--help`, since creation is flagged as the sensitive operation.
Trim the installation section to the two most common paths (npm and the standalone installer) and mention the others in one line.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dominated by executable commands with no explanations of concepts Claude already knows, which is good, but when-to-use guidance is stated three times ("Important expectations", "When to use it", "Pitfalls"), the Core rule set is largely restated in Pitfalls, and five install paths are enumerated. This fits anchor 3 ("mostly efficient but... could be tightened") better than 4, where over-explanation would be minor only. | 3 / 5 |
Actionability | Nearly every section gives copy-paste-ready commands with realistic flags and example queries, covering search, extract, research, enrich, and findall. It falls just short of anchor 5 because `monitor create` defers entirely to `--help` with no concrete invocation and `PARALLEL_API_KEY="***"` is a placeholder, leaving minor gaps. | 4 / 5 |
Workflow Clarity | Research and enrichment workflows are explicitly sequenced (launch with --no-wait, capture ID, poll, validate output before reporting success) and the error-handling section pairs exit codes with an auth-recovery loop (check auth, confirm key, verify PATH). This matches anchor 4 ("clear sequence with most checkpoints present; minor validation gaps") — research/findall result handling lacks explicit validation checkpoints, so it is not 5. | 4 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), and the full per-subcommand CLI reference (FindAll, Monitor, Enrichment details, processor tiers) is inlined in a single ~380-line SKILL.md. Section headers are consistent so navigation is possible, but content that clearly belongs in one-level-deep reference files is inline, matching anchor 3 rather than 2 (structure is good, not minimal) or 4 (nothing is split out at all). | 3 / 5 |
Total | 14 / 20 Passed |