Content
88%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 a highly actionable, well-sequenced diagnostic workflow: platform detection, an ordered nine-item checklist with specific MCP tool calls and parameters, phased dashboard checks with checkbox lists, and an explicit verify-the-fix section with feedback loops. Conciseness and progressive disclosure are good but not excellent — around 100 lines of reference tables are inlined that could live in separate files, and a few 'common issues' entries restate checklist content.
Suggestions
Move the trailing 'Reference:' sections (SDK error codes, known platform issues, platform configuration checklists) into separate files such as references/error-codes.md and references/known-issues.md, keeping only the highest-frequency entries inline so SKILL.md remains a workflow overview.
Trim the 'Reference: common issues' section — entries like 'Offering returns empty' and 'User purchased but has no entitlement' restate universal checklist steps 3–5 and the Phase B checks.
Consolidate the repeated pacing guidance ('Work them in order. Most reports resolve before you reach the platform specifics.' and 'Most reports are resolved by steps 1 through 5') into a single statement.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with RevenueCat-specific knowledge Claude would not have (MCP tool parameters, iOS 18 StoreKit bugs, exact sandbox-account settings paths) and never explains generic concepts, so it is largely efficient. Minor trimming is possible: 'Reference: common issues' entries like 'Offering returns empty — Verify a current offering exists, packages have products attached' restate checklist steps 3–5, and the pacing note ('Most reports resolve before you reach the platform specifics' / 'Most reports are resolved by steps 1 through 5') appears twice — anchor 4, not 5. | 4 / 5 |
Actionability | Fully executable throughout: named MCP tools with parameters ('list-offerings … passing limit: 100 and expand=items.package.product'), CLI mappings ('list-offerings → rc offerings list'), runnable snippets ('Purchases.shared.getCustomerInfo(fetchPolicy: .fetchCurrent)'), and exact settings paths ('Settings → Developer → Sandbox Apple Account on iOS 18 and later'). Copy-paste-ready guidance covers the common cases, matching anchor 5. | 5 / 5 |
Workflow Clarity | Clear five-stage sequence (detect platform → ordered nine-item checklist → phased dashboard checks → platform specifics → verify) with explicit ordering ('Walk these nine items in order'), checkbox checklists in Phase B, and a dedicated validation section: 'Do not declare the issue fixed until: 1. The log … shows the expected success line … 2. The dashboard reflects the change … 3. The reproduction steps … now pass', plus error recovery ('If the user cannot reproduce locally, have them send the full debug log'). Matches anchor 5; the destructive-operation cap does not apply because fixes require user confirmation ('confirm with the user, then execute via MCP') and are verified in section 5. | 5 / 5 |
Progressive Disclosure | Platform specifics are correctly split into clearly signaled, one-level-deep files ('Read the platform file that matches detection … platforms/ios.md' … 'platforms/react-native.md'), and related skills are referenced by name. However, roughly 100 lines of trailing 'Reference:' sections (SDK error-code tables, known platform issues, platform configuration checklists) are inlined in SKILL.md when they could be separate reference files, and the iOS/Android configuration checklists overlap with the platform-file layer — anchor 4's 'minor organization gaps', short of anchor 5's fully split content; better than anchor 3 because references are clearly signaled and the workflow overview is appropriately placed. No bundle files were provided, so the referenced platforms/*.md files could not be verified against the bundle. | 4 / 5 |
Total | 18 / 20 Passed |