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.
This is an exceptionally strong instruction-only skill body: every phase is sequenced with explicit read-before-report validation, the grep patterns and commands are copy-paste ready, and the 20 anti-patterns each carry measured specifics, exceptions, and concrete fixes rather than generic advice. The main improvements are removing the duplicated false-positive list (inline exceptions vs. the closing "False Positives" section) and considering whether the pattern catalog belongs in a reference file to keep the main body lean.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with measured, non-obvious facts ("measured on Xcode 27.3", "in testing on one app and device") and wastes no tokens explaining concepts Claude already knows. It stays at 4 rather than 5 because the false-positive exceptions (`.system.searchInApp`, `@Dependency(default:)`, schema-entity plain `var`s, stored `UUID` ids) are stated twice — inline in Phase 2 patterns and again in the "False Positives" section — and the four near-identical Invocation Examples could be cut to one. | 4 / 5 |
Actionability | The skill provides copy-paste-ready Grep regexes ("`import\s+AppIntents`", "`@AppIntent\(\s*\[?\.tool`"), an exact verification command ("`xcrun sirimetadataanalyzer <App.app> --format json`"), per-pattern severity/confidence labels with concrete fixes ("`some IntentResult & ReturnsValue<T>`", "throw `AppIntentError.UserActionRequired.signin`"), and a complete output template. This fully matches anchor 5 ("Fully executable; copy-paste ready code or commands; specific examples cover the common cases"). | 5 / 5 |
Workflow Clarity | Five explicitly sequenced phases each define their outputs, with validation checkpoints throughout: "Confirm every candidate finding by reading the surrounding code before reporting it", "Each finding needs a file:line you read", "report 'bundle membership unverified' instead of a finding", plus a terminal verify chain (build → sirimetadataanalyzer → SiriEvaluations → device) and overflow rules for >50/>100 findings. This matches anchor 5 ("Clear sequence with explicit validation steps; feedback loops for error recovery"); it is a read-only audit, so the destructive-operation cap does not apply. | 5 / 5 |
Progressive Disclosure | No bundle files exist, and the body is well-organized with clear section headers and a "Related" section that cleanly signals one-level-deep external pointers ("`axiom-integration (skills/app-intents.md)`", "`security-privacy-scanner` agent"). It stays at 4 rather than 5 because the ~140-line, 20-pattern catalog is fully inlined — the body is the complete detailed material rather than an overview pointing to reference files, which the rubric's rationale favors splitting. | 4 / 5 |
Total | 18 / 20 Passed |