Content
67%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-organized, safety-conscious orchestration skill body: lean bullets, defined modes, platform-specific rules, and a concrete output template, with review-first gating properly wired into the workflow. Its main weakness is actionability — the scoring model and several rules stay qualitative without weights, thresholds, or examples that would make execution deterministic.
Suggestions
Add concrete weights or a scoring heuristic to the Scoring Model (e.g., point values per positive/negative signal and a confidence threshold for the prune queue) so 'score prune candidates' becomes executable rather than qualitative.
Deduplicate the safety rules: consolidate the repeated no-auto-send and LinkedIn review-first statements between Safety Defaults, Platform Rules, and Outbound Rules into one authoritative section.
Add a post-review feedback step to the workflow (e.g., what to do when the user rejects or modifies queue items) to close the validation loop before the apply step.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is almost entirely lean bullet lists with no explanation of concepts Claude already knows, but there is minor redundancy: "do not auto-send" in Safety Defaults is repeated in Outbound Rules, and the LinkedIn review-first rule appears in both Safety Defaults and Platform Rules. | 4 / 5 |
Actionability | Concrete elements exist (named tools like x-api, lead-intelligence, brand-voice; three defined modes; a full review-pack output template), but the core Scoring Model is qualitative — signals like "reciprocity" and "low-value noise" are listed with no weights, thresholds, or worked examples, and there is no executable guidance for the key steps. | 3 / 5 |
Workflow Clarity | The 8-step workflow is clearly sequenced and the destructive/batch nature is handled with explicit review gating ("Return a review pack before any apply step", "still review-gated before apply"), so the batch-operation cap does not bind. The minor gap is the absence of a feedback loop for when the user rejects or edits queue items. | 4 / 5 |
Progressive Disclosure | The single-file body is well-sectioned (When to Activate, Required Inputs, Modes, Platform Rules, Workflow, Review Pack Format) with appropriately overview-level content and clear navigation, though it is a monolithic file with no reference files despite listing several related skills. | 4 / 5 |
Total | 15 / 20 Passed |