Content
85%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.
A well-designed router: concrete commands, unambiguous routing logic with redundant decision aids, and clearly signaled one-level-deep references. Its two real weaknesses are redundant routing representations (the dot tree and example invocations restate the triggers) and the fact that all four referenced sub-skill files are absent from the bundle, making the progressive-disclosure chain unverifiable.
Suggestions
Ship (or correct the paths to) the four referenced files — skills/xcode-mcp-setup.md, skills/xcode-mcp-tools.md, skills/xcode-mcp-ref.md, and skills/axe-ref.md — so the 'Read:' directives resolve; every pointer currently dangles.
Cut one of the redundant routing representations: the dot decision tree or the Example Invocations section restates the Routing Logic triggers almost verbatim.
Move the MCP-vs-CLI capability comparison (xcui/xclog/xcsym/xcprof details) into skills/xcode-mcp-tools.md; it is tool-choice guidance, not router-level content. Also disambiguate the 'axiom-tools (skills/device-control-ref.md)' reference to state which bundle owns that path.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient: the version-gated guidance (Xcode 26.x requires a running Xcode vs. 27's `xcrun mcp-server`) and the anti-rationalization/conflict tables are non-obvious, earned tokens. But the dot decision tree, the "When to Use" list, the Routing Logic triggers, and the Example Invocations restate the same routing decision four times — the dot graph and parts of the example list could be trimmed. Fits anchor 4 (efficient, minor trimming possible), not 5. | 4 / 5 |
Actionability | Fully executable guidance: copy-paste commands (`sudo xcrun mcp-server enable`, `xcrun mcp-server open <project-or-workspace-path>`, `command -v <tool>`, `MCP_XCODE_PID`, `xcrun mcpbridge run-agent`), an unambiguous plugin-root resolution rule ("two levels above this skill's Base directory"), and explicit "Read: skills/..." directives per trigger. Example invocations cover the common cases including the combined tools-then-ref case. | 5 / 5 |
Workflow Clarity | As a router, the single action — match the question to a sub-skill and read it — is unambiguous, reinforced consistently by triggers, the decision tree, the anti-rationalization table, and the conflict-resolution ownership table, with a demonstrated chaining pattern for combined questions ("Read: skills/xcode-mcp-tools.md (workflow), then skills/xcode-mcp-ref.md (params)"). No destructive or batch operations are executed in this file, so no validation cap applies. | 5 / 5 |
Progressive Disclosure | The written structure is exemplary — one-level-deep references, each with an explicit "Read:" signal, plus a Resources section listing all four files. However, none of the referenced files (skills/xcode-mcp-setup.md, skills/xcode-mcp-tools.md, skills/xcode-mcp-ref.md, skills/axe-ref.md) exist in the bundle, so every navigation pointer dangles, and the reference "axiom-tools (skills/device-control-ref.md)" is ambiguous about which skill owns that path. Good signaling cannot score above 3 when the disclosure chain leads to files that are not present. | 3 / 5 |
Total | 17 / 20 Passed |