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 delivers highly actionable, safety-first operational rules with clearly sequenced workflows and approval gates, and it wastes little space on known concepts. Its main weaknesses are redundancy across the Safety Rules, Credit-Aware, and Mutation sections, a monolithic structure with no reference files despite its length, and missing error-recovery loops.
Suggestions
Consolidate the redundant approval/credit rules: fold the Safety Rules section's restatements of Credit-Aware Behavior and Mutations gates into those sections, or reduce Safety Rules to a short cross-referencing checklist, to cut repeated tokens.
Split the per-domain rule sets (e.g., output table specifications, sequence/launch rules, mutation gates) into references/ files with clearly signaled one-level-deep links from SKILL.md, keeping the body as a lean overview.
Add brief error-recovery feedback loops for batch/destructive operations — e.g., what to do when a mutation fails, a duplicate is detected, or dedupe is unsupported — to complement the existing preview-and-confirm checkpoints.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and rule-driven with no explanation of concepts Claude already knows, but the Safety Rules section substantially restates the credit/approval gates from "Credit-Aware Behavior" and "Mutations", and Output Rules repeat the "Apollo credit" phrasing — several passages could be tightened or merged. | 3 / 5 |
Actionability | Concrete, executable guidance throughout an instruction-only skill: an exact 7-step People-to-Company-Details flow, exact table column specifications, mandated phrasing ("Say 'Company Details,' not 'account context.'"), and field-level diff requirements. It stops short of a 5 because there are few worked examples beyond the single output template. | 4 / 5 |
Workflow Clarity | Multi-step flows are clearly sequenced with explicit approval checkpoints before mutations and enrollment (preview, diff, explicit confirmation, "Do not split a risky batch into multiple mutations to bypass review"), so the destructive/batch cap does not apply. However, error-recovery feedback loops (what to do when a mutation fails or duplicates are flagged) are absent, keeping it below a 5. | 4 / 5 |
Progressive Disclosure | No bundle files exist and the ~213-line body keeps every per-domain rule set inline (output table specs, sequence rules, mutation gates) that could reasonably live in separate reference files. Section headers provide real structure, so it is above a 2, but there are no well-signaled internal references — the only pointers target a parent skill outside this bundle, one with a malformed anchor ("plugins/sales/skills/index/SKILL.md## Cross-Skill Best Practices"). | 3 / 5 |
Total | 14 / 20 Passed |