Content
71%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, domain-rich instructional skill with concrete formulas, thresholds, worked examples, a sequenced decision framework, and clean one-level progressive disclosure. Its main weakness is editorial prose that pads the length without adding actionable content.
Suggestions
Trim motivational/editorial paragraphs (e.g., the opening 'moment of truth' paragraph and the Closing section) to lean prose that assumes the reader's competence and keeps every token actionable.
Pull one concrete setup checklist for the highest-value test (geo holdout: matched markets, power calculation, pre-committed window) inline from the reference, so the core causal method is executable without a file hop.
Add an explicit reconciliation validation checkpoint to the 12-consideration framework (e.g., 'recompute blended CAC from warehouse totals and confirm it does not rely on platform sums') so the workflow has a visible verify step.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient domain expertise (attribution, reconciliation, incrementality) that Claude does not already know, but it carries several paragraphs of editorial/philosophical prose ('The dashboard is the moment of truth...', 'The platform will not tell you. The platform's incentive is the opposite.') that could be tightened without losing substance. | 3 / 5 |
Actionability | Provides concrete, specific guidance — the blended CAC formula, detection thresholds ('exceed warehouse-attributed... by more than 30%'), incremental rate ranges by channel, and worked ROAS/LTV examples — though some concrete 'how' steps (e.g., geo holdout setup) are deferred to references rather than fully spelled out. | 4 / 5 |
Workflow Clarity | A clear sequenced checklist (the 12 considerations) with an explicit decision output (Scale/Hold/Kill) and a sanctioned 'state the gap' checkpoint for missing data, but no explicit validate-fix-retry feedback loop, which is appropriate here since the skill is interpretive rather than a destructive/batch operation. | 4 / 5 |
Progressive Disclosure | Clear overview with seven well-signaled one-level-deep references, each linked inline at the relevant section and consolidated in a Reference files list; all referenced files exist on disk and content is appropriately split between inline concepts and detailed reference playbooks. | 5 / 5 |
Total | 16 / 20 Passed |