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 is well-structured and actionable with clear tool sequences and a real async feedback loop for destructive batch operations. Its main weaknesses are duplicated pitfalls and generic boilerplate that hurt conciseness, plus all reference material being inlined rather than split across bundle files.
Suggestions
Remove the 'Known Pitfalls' section or collapse it into the per-workflow 'Pitfalls' to eliminate duplication, and delete the generic 'When to Use'/'Limitations' boilerplate that adds no Amplitude-specific value.
Move the Quick Reference table and Common Patterns into a references/ file (e.g. REFERENCE.md) with a clearly signaled one-level-deep link from the overview to improve progressive disclosure.
Replace 'check schema for maximum events per request' with the concrete batch limit, or at minimum document the known limit, to close the actionability gap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The per-workflow sections are reasonably tight, but the 'Known Pitfalls' section duplicates pitfalls already stated in each workflow (user IDs, timestamps, async cohort), and the generic 'When to Use'/'Limitations' boilerplate ('Do not treat the output as a substitute for environment-specific validation...', 'Stop and ask for clarification...') adds no Amplitude-specific knowledge Claude lacks; it could be tightened. | 3 / 5 |
Actionability | Provides concrete tool slugs, numbered tool sequences, key parameters, and an executable user_properties JSON example; minor gaps where it defers specifics to the schema ('check schema for maximum events per request') keep it just below 5. | 4 / 5 |
Workflow Clarity | Workflows are clearly numbered with prerequisites, a setup validation checkpoint ('Confirm connection status shows ACTIVE before running'), and an explicit async feedback loop (CHECK_COHORT_STATUS repeated until 'complete'/'error') for the destructive batch cohort operation; the batch event path lacks a landing-verification step, so not a 5. | 4 / 5 |
Progressive Disclosure | Sections are well-organized, but reference-type material (Quick Reference table, Common Patterns, Known Pitfalls) is all inlined in a single ~220-line SKILL.md with no bundle files in references/scripts/assets and no one-level-deep external references to split it out. | 3 / 5 |
Total | 14 / 20 Passed |