Content
100%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 exemplary reference-type skill body: all MPS-specific gotchas are surfaced as terse Critical Directives, the workflow carries genuine before/after validation gates with a rebuild-reload loop and scratch-model cleanup, and detail is correctly pushed to one-level-deep reference files with task-based routing. The body teaches only what Claude could not already know and provides exact MCP tool invocations with parameters. No dimension falls short of its top anchor.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body explains nothing Claude already knows — every line is an MPS-specific gotcha ("the `infer` prefix makes an inequation *soft* … will not immediately error", "**always wrap the real Expression in a `NormalTypeClause`**", "Quick fixes … are roots, **not** executed automatically") or a pointer. Matches anchor 5 ("Lean and efficient; assumes Claude's competence; every token earns its place"); the only near-duplication (the `List<SNode>` → `sequence<node<X>>` pointer appearing in both Critical Directives and Related Skills) is cross-referencing in two navigational contexts, not padding. | 5 / 5 |
Actionability | Gives copy-adaptable commands with exact parameters: "`mps_mcp_create_model` with `moduleName: \"<lang>\"` and `modelName: \"<lang>.typesystem\"`", "`mps_mcp_apply_intention(nodeReference = <problem node>, intentionId = <fix id>)`", "MAKE with `parameters` = `{\"modules\":[\"<lang>\"],\"rebuild\":true}`", and the executable statement form `error "Expected " + <expectedType> + " but got " + <actualType> -> node;`. Matches anchor 5 ("Fully executable; copy-paste ready code or commands; specific examples cover the common cases") — common cases are routed to concrete reference files rather than left abstract. | 5 / 5 |
Workflow Clarity | The 0–9 workflow has explicit validation checkpoints and a feedback loop: step 3 takes a baseline ("**Before** the change: `mps_mcp_check_root_node_problems` + `EXPLAIN_TYPE` … so you have the baseline error"), step 5 re-verifies after ("**After**: ask again — `IS_SUBTYPE` … `EXPLAIN_TYPE`"), step 8 validates, and step 9.6 cleans up the scratch model. Matches anchor 5 ("Clear sequence with explicit validation steps; feedback loops for error recovery"), not 4 — the checkpoints are explicit before/after gates, not implicit. | 5 / 5 |
Progressive Disclosure | The Reference Index opens with task-based routing ("**Start here — most common case**: one `InferenceRule` that types a concept → read only `references/inference-rules.md` …") and each reference entry states both what it contains and when to open it ("Open `references/when-concrete.md` when a rule must wait for a resolved concrete type"). References are one level deep and clearly signaled, matching anchor 5 ("Clear overview with well-signaled one-level-deep references; content appropriately split; easy navigation"). No bundle files were provided for structural verification, so this is scored on the body's structure; the only wrinkle — the cross-skill `type-queries.md` pointer appearing inline rather than in the index — is signaled at its point of use. | 5 / 5 |
Total | 20 / 20 Passed |