Content
57%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 rich, domain-knowledge-heavy body with genuinely specific CRO data and an explicit output format, but it is long and partially redundant, reads as a reference manual rather than a sequenced workflow, and keeps everything inline instead of splitting stable detail into reference files. The description is in much stronger shape than the body.
Suggestions
Merge the "Task-Specific Questions" section into "Initial Assessment" and cut bullets that restate common knowledge (easy dismiss, don't trap users, basic accessibility tenets) to reduce token spend.
Frame the body as an explicit ordered workflow (assess context, select type and trigger, draft copy, set frequency and compliance rules, produce the Output Format deliverable) with a final check that compliance and frequency rules are covered before emitting the recommendation.
Move the Experiment Ideas and Common Popup Strategies sections into one-level-deep reference files (e.g., references/experiments.md, references/strategies.md) and link them from a lean overview, since this material is lookup material rather than needed on every invocation.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly dense, useful bullets (benchmarks, copy formulas, trigger numbers), but the ~450-line body includes redundant question lists ("Initial Assessment" bullets vs the "Task-Specific Questions" section), ~90 lines of brainstorm experiment bullets, and reminders Claude already knows ("Easy to dismiss", "Don't trap or trick", generic accessibility tenets). Not 2 because the bulk is genuine domain expertise rather than padded explanation. | 3 / 5 |
Actionability | Concrete guidance throughout: copy templates ("Get [result] in [timeframe]", decline options), specific numbers (30-60s delay, 25-50% scroll depth, 400-600px width, 2-5% benchmarks), and an explicit Output Format defining the deliverable. Not 5 because several sections stay abstract ("Address common objections", "Respect user choice") without worked examples. | 4 / 5 |
Workflow Clarity | A clear starting step (read product-marketing context, assess purpose/state/traffic) and an implicit sequence (assess, choose trigger/type, design, copy, frequency, compliance, output) exist, but the process is never framed as an ordered procedure and there are no validation checkpoints (e.g., confirm compliance before finalizing the recommendation). Not 2 because the Initial Assessment section does sequence the intake; not 4 because checkpoints are absent. | 3 / 5 |
Progressive Disclosure | No bundle files exist, so the entire knowledge base (experiment ideas, per-industry strategies, type-by-type playbooks) is inlined in a single ~450-line SKILL.md. Section headers and separators keep it navigable, but content that clearly belongs in one-level-deep reference files is inline, with no references to signal. | 3 / 5 |
Total | 13 / 20 Passed |