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.
A well-engineered reference-hub body: the section-index table, critical directives, and the multi-valued-reference walkthrough deliver genuinely non-obvious MPS guidance with one executable blueprint and explicit verification steps. The main room for improvement is tightening the companion-loading preamble and adding one minimal inline CREATE_CONCEPTS blueprint so the most common operation is executable without a file read.
Suggestions
Tighten the "Loading companion skills" section — its four conditional clauses can be compressed to one or two sentences, freeing tokens for the operations themselves.
Inline a minimal two-concept `CREATE_CONCEPTS` blueprint (with one property and one child) so the most common greenfield case is executable directly from SKILL.md before opening the reference file.
Add a brief verify step after bulk `CREATE_CONCEPTS` calls (e.g. re-query the structure model or check `makeStatus`) to close the last validation gap in the workflow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and almost entirely devoted to MPS-specific rules Claude would not know (smart-reference criteria, single-valued reference links, alias guidance), but the "Loading companion skills" paragraph and some restatement between Critical Directives and Workflow steps could be trimmed — anchor 4 (efficient, minor trims), not 5 because of those, not 3 since nothing explains generally known concepts. | 4 / 5 |
Actionability | Concrete tool calls (`mps_mcp_get_project_structure`, `mps_mcp_alter_nodes` with `MAKE` and `rebuild="true"`, `mps_mcp_query_structure IS_SMART_REFERENCE` with expected output) and a copy-paste-ready JSON blueprint for the wrapper pattern are present, but exact JSON parameter formats for the primary create/update flows are deferred to reference files — anchor 4 (mostly executable, minor gaps), not 5 given the deferred formats, not 3 since the inline example is fully executable. | 4 / 5 |
Workflow Clarity | An 11-step numbered sequence runs from locating the language to a mandatory rebuild, with explicit validation checkpoints ("Confirm with `mps_mcp_query_structure` `IS_SMART_REFERENCE` on the wrapper: expect `isSmartReference: true`"); however, bulk `CREATE_CONCEPTS` operations lack an explicit verify/retry loop — anchor 4 (clear sequence, minor validation gap), not 5 because of that gap, not 3 since validation exists for the risky smart-reference path. | 4 / 5 |
Progressive Disclosure | Navigation is exemplary: a "Section index — one row, one read" table maps each job to exactly the files to read, a "Start here — most common case" pointer orients the reader, and all references are one level deep (`references/structure-operation-api/*.md`, `references/attributes-and-annotations.md`) with the body acting as a lean hub — anchor 5 (clear overview, well-signaled one-level-deep references, appropriately split content). | 5 / 5 |
Total | 17 / 20 Passed |