Content
67%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 tightly written, expert-facing reference: concrete MCP tool invocations with parameters, a high-value modifier table, and a validated workflow. Its one serious defect is that the entire progressive-disclosure layer it is built around — the nine reference files it routes to — is absent from the bundle, leaving the detailed blueprints, examples, and troubleshooting unreachable.
Suggestions
Ship the `references/` directory: every file named in the Reference Index (method-declarations.md, json-blueprints.md, local-and-super-calls.md, inheritance-and-dispatch.md, constructors.md, calling-from-other-aspects.md, lang-core-behavior-overrides.md, common-failures.md, aspect-model-stereotypes.md) is currently missing, so the 'Open X when...' routing leads nowhere.
Inline the minimal `ConceptBehavior` / `ConceptMethodDeclaration` JSON blueprint (currently deferred to the missing json-blueprints.md) so the most common case is executable from SKILL.md alone, and note the `List<SNode>` → `sequence<node<X>>` return-type fix once instead of three times.
Add an explicit error-recovery loop to the workflow (run `mps_mcp_check_root_node_problems` → fix reported problems → re-check before rebuild) rather than delegating failure handling to the absent common-failures.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and expert-level with no padding — the modifier table and Critical Directives carry high information per token and assume MPS competence ("the `constructor` child is **mandatory**", "abstract implies virtual"). Not 5 because there is minor repetition: the MPS-typed return-type / Java-parser caveat appears three times (Critical Directives, workflow step 4, Related Skills) and null-return behavior twice. Not 3 because these are trimmable instances, not whole padded sections explaining things Claude already knows. | 4 / 5 |
Actionability | Guidance is concrete and executable: named MCP tools with exact parameters (`mps_mcp_create_model` with `moduleName`/`modelName`, `mps_mcp_parse_java_and_insert` with `featureKind: "METHOD"`, `insert: {mode: "child", parentRef: ..., role: "method"}`), plus concrete defaults ("empty body is fine"). Not 5 because the actual JSON blueprints users would copy-paste are deferred to `references/json-blueprints.md` — which is missing from the bundle — so the common case cannot be completed end-to-end from the body alone. Not 3 because what is inline is specific, not pseudocode. | 4 / 5 |
Workflow Clarity | The Common-Path Workflow is a clear 1–5 sequence with an explicit validation checkpoint (step 5: `mps_mcp_check_root_node_problems`, then rebuild the language), reinforced by the Critical Directive to run it after edits. Not 5 because the body has no explicit error-recovery loop (validate → fix → re-validate) — failure handling is delegated to a `references/common-failures.md` file that does not exist in the bundle. Not 3 because validation is explicit and repeated, not merely implicit. | 4 / 5 |
Progressive Disclosure | The body's structure is exemplary — a 'Start here' routing paragraph plus per-reference 'Open X when...' triggers, all one level deep — but scored against the actual bundle, the disclosure layer is broken: there is no `references/` directory, so all nine-plus referenced paths (`method-declarations.md`, `json-blueprints.md`, `inheritance-and-dispatch.md`, `constructors.md`, `lang-core-behavior-overrides.md`, `common-failures.md`, etc.) are dangling and none of the detailed content can actually be loaded. Not 3/4 because the rubric's structural anchors (clear signals, one level deep) are met in the text but not in the bundle — navigation leads nowhere, which is a worse failure than unclear signaling. | 2 / 5 |
Total | 14 / 20 Passed |