Content
71%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.
An information-dense, highly actionable router skill with exemplary progressive disclosure — every routing decision names a specific file, API, or error, and detail is uniformly delegated to clearly-labeled one-level-deep references. The main cost is token efficiency: the same routing knowledge is repeated in up to four different sections, and one placeholder section (Cursor Subagent Routing) is too vague to act on.
Suggestions
Consolidate the redundant routing sections: the Decision Tree, Example Invocations, and "Critical Patterns" largely restate the Approach Triage table and Routing Logic — keep the triage table plus one canonical decision tree and trim the rest to reclaim context-window budget.
Either flesh out the "Cursor Subagent Routing" section (name the available subagents and when each is called) or delete it — "Delegate to the appropriate Cursor subagent" with no named options is the body's only unactionable instruction.
Add a terminal fallback branch to the Decision Tree (what to do when a query matches no entry) to close the last workflow-clarity gap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every line carries non-obvious, version-specific Apple facts (kappa thresholds, entitlement names, per-OS adapter pinning) rather than concepts Claude already knows, but the same routing decisions are restated across four sections — the Approach Triage table, Routing Logic, the Decision Tree, and Example Invocations — and the "Critical Patterns" section near-duplicates the Foundation Models Work listing. Could be tightened by consolidating; not 2 because there is no padding or generic explanation. | 3 / 5 |
Actionability | Routing is concrete throughout: named files to read per situation, named APIs, error cases, entitlements, and ~20 User-query → Read `skills/X.md` example invocations covering the common cases. Not 5 because of the near-empty "Cursor Subagent Routing" section — "Delegate to the appropriate Cursor subagent when this router calls for a specialized auditor" names no subagents beyond foundation-models-auditor, leaving that instruction un-executable as written. | 4 / 5 |
Workflow Clarity | The triage workflow is clearly sequenced ("First, determine which kind of AI the developer needs" → intent table → boundary clarifications → decision tree) and includes a clarification checkpoint ("First ask: Apple Intelligence or custom ML model?") plus an Anti-Rationalization table guarding wrong turns. Not 5 because the routing process itself has no verification/feedback loop (e.g., what to do when a query matches none of the tree entries or when a routed file is missing). | 4 / 5 |
Progressive Disclosure | A pure router/overview: virtually all detail is delegated to one-level-deep, clearly signaled `skills/*.md` references, each annotated with what it contains, and navigation is served three ways (triage table, decision tree, example invocations). No reference material is inlined in the body; only two external references are marked "See `claude-api` skill (external)" so nesting never goes past one level. | 5 / 5 |
Total | 16 / 20 Passed |