Content
73%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-structured instruction skill with an explicit, well-gated workflow, concrete match/de-dup formulas, and clear boundaries against neighboring skills. Its main weakness is redundancy — the handoff and truth-set rules are restated several times — and cross-skill references with deep relative paths that may not resolve outside the parent repo.
Suggestions
State the roi-calculator / ad-account-auditor delegation once (e.g. in the intro blockquote or Next Best Skill) and trim its repetition from the Skill Contract, Step 7, and Reference Materials to cut ~15% of tokens.
Consolidate the repeated 'order IDs are the truth set, never platform-reported counts' rule into a single highlighted statement instead of restating it in the blockquote, the body paragraph, and Steps 1/3/4.
Move the inline cross-skill links scattered through the body into the single Reference Materials section, and verify the deep relative paths (../../../references/..., ../../orchestrate/...) resolve from the skill's install location or shorten them.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient with no basic-concept tutoring, but noticeably padded by repetition: the delegation to roi-calculator/ad-account-auditor is restated in the intro blockquote, the Skill Contract ('Done when' bullet), Step 7, Reference Materials, and Next Best Skill, and the order-ID truth-set rule is repeated three times. This fits anchor 3 ('mostly efficient but... could be tightened') better than anchor 4, where over-explanation would be only minor. | 3 / 5 |
Actionability | Concrete, executable guidance throughout: explicit match labels ('matched', 'double-counted', 'unmatched'), join keys ('order ID (preferred) or timestamp + value as a fallback'), computed formulas ('double-count rate (claimed conversions / real orders)', 'incremental orders ÷ exposed'), a literal output path, and a concrete failure status ('return status: NEEDS_INPUT'). As an instruction-only skill the absence of code is not penalized; it falls short of anchor 5 only because some steps (e.g. 'convert all monetary values to one currency at a stated rate', model comparison 'e.g. last-click vs linear') leave specifics to the model's discretion. | 4 / 5 |
Workflow Clarity | The 7 numbered steps are clearly sequenced (confirm truth set → normalize → match → de-dup → compare models → incrementality → hand off) with an explicit validation gate in Step 1 ('if it is absent, return status: NEEDS_INPUT... do not reconcile'), error-recovery paths (unresolved mismatches pushed to open-loops.md), edge-case branches (no holdout → 'mark incrementality N/A'), and a 'Done when' checklist — matching anchor 5's validation-plus-checklist profile. | 5 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are absent), so this scores the body's own structure: well-organized sections with a dedicated 'Reference Materials' list where each link carries a one-line purpose. It is not anchor 5 because many references use fragile deep relative paths outside the skill directory (e.g. '../../../references/roas-benchmark.md', '../../orchestrate/ad-test-designer/references/measurement-control.md') that cannot be verified to resolve, and the same links are repeated inline throughout the body rather than consolidated. | 4 / 5 |
Total | 16 / 20 Passed |