Content
78%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 well-structured entrypoint with concrete API-level decisions, mistakes, and a review checklist, backed by an appropriately split one-level-deep reference bundle. Its main weaknesses are the triplicated reference routing (a token-efficiency cost) and the absence of any complete executable snippet or error-recovery loop in the entrypoint itself.
Suggestions
Consolidate the reference routing: the intro paragraph, "Route by Task", and "References" all list the same three files — merge into one routing section (e.g., keep "Route by Task" and drop the duplicate links in the intro and References list) to cut redundant tokens.
Add one minimal copy-paste-ready snippet in the entrypoint for the most common case (e.g., a complete SwiftUI Map + liveUpdates task cancellation pattern) so the highest-frequency workflow is executable without opening a reference.
Include a brief error-recovery loop in the workflow (e.g., what to do when authorization is denied mid-session or a route request fails and when to retry) to complement the existing verification step and checklist.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence (no basic MapKit/CLLocation explanations), but the same three reference files are routed three separate times (intro paragraph, "Route by Task", and "References"), which is trimmable redundancy. It sits above the mostly-efficient anchor at 3 but short of "every token earns its place" at 5. | 4 / 5 |
Actionability | Concrete, API-level guidance throughout — exact names like "CLServiceSession", "CLBackgroundActivitySession", a copy-paste debounce recipe (".task(id: searchText) + Task.sleep(for: .milliseconds(300))"), "placemark.locality ?? \"Unknown\"", the 20-condition CLMonitor limit, and "NSLocationWhenInUseUsageDescription" — but the entrypoint contains no complete executable example, deferring all full code to references. This is mostly-executable guidance with minor gaps rather than fully copy-paste-ready coverage. | 4 / 5 |
Workflow Clarity | The five-step workflow is clearly sequenced and ends with an explicit verification step ("Verify empty results, movement, camera changes, route failures, localization, offline/network transitions, and background rules") reinforced by a full Review Checklist. It falls short of anchor 5 because the workflow itself lacks error-recovery feedback loops (validate → fix → retry), though the build/review nature of the skill means no destructive-operation cap applies. | 4 / 5 |
Progressive Disclosure | The ~87-line body is a genuine overview: it stays at the decision/checklist level while all implementation detail lives in three real, one-level-deep reference files (verified to exist, ~400–780 lines each, with no nested .md chains beyond sibling section links). Each reference is well-signaled with what it covers and when to read it, matching the clear-overview anchor exactly. | 5 / 5 |
Total | 17 / 20 Passed |