Content
92%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 an exemplary reference-skill overview: critical directives and a validated six-step workflow inline, with dense, verifiable MPS specifics (concept UUIDs, MCP tool signatures, action ids) and a well-signaled one-level-deep reference index. The only weakness is minor redundancy — failure-diagnosis detail repeated between Critical Directives, workflow step 6, and the common-failures reference. Actionability and workflow clarity are top-tier for an instruction-only skill.
Suggestions
Compress the deep failure-diagnosis detail in the fail-first directive (e.g. the '`Node: null vs <root>`' / `NO_GENERATORS` interpretation and `AssumptionViolatedException` causes) into a one-line pointer to `references/common-failures.md`, which already covers them.
State the `AssumptionViolatedException` skipped-vs-failed reporting nuance once (either Critical Directives or workflow step 6) and reference the other location instead of repeating it.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is information-dense MPS-specific knowledge Claude would not have (concept UUIDs, MCP tool parameters, platform action ids, cell-id conventions) with essentially no explanation of already-known concepts, so it sits above anchor 3's 'some unnecessary explanation'. It falls short of anchor 5's 'every token earns its place' due to a few redundant spots: the `AssumptionViolatedException` skipped-vs-failed nuance appears in both the fail-first directive and workflow step 6, and deep failure-diagnosis detail like "`Node: null vs <root>` means no generator ran (`mps_mcp_generation` `PLAN` shows `NO_GENERATORS`…)" duplicates what `references/common-failures.md` covers. | 4 / 5 |
Actionability | Guidance is copy-paste concrete throughout: exact MCP calls with parameters ("`mps_mcp_create_module(type="solution", …, facets="[\"tests\"]")`" with the self-describing JSON response), a concrete concept ref ("instantiate `SimpleNodeTest` (`c:8585453e-6bfb-4d80-98de-b16074f1d86c/1225978065297`)"), exact platform action ids ("`invoke action by id: $Undo` (`$Redo`, `$Copy`, `$Cut`, `$Paste`)") and cell-id conventions ("target `empty_<role>` for the caret — *not* `refNodeList_<role>`"). Every workflow step names the specific tool to use, matching anchor 5. | 5 / 5 |
Workflow Clarity | The Common-Path Workflow is a clearly sequenced six-step process with an explicit validation checkpoint before running ("**Validate** with `mps_mcp_check_root_node_problems` on the new root") and concrete run instructions including how to read the output. Feedback loops are explicit: "A test written first must fail first, for the reason the change addresses… fix it and run again", plus failure-triage guidance for both editor and generator tests. This matches anchor 5's sequence + explicit validation + error-recovery loops. | 5 / 5 |
Progressive Disclosure | The body stays an overview: critical directives and the common-path workflow inline, with a Reference Index that signals exactly when to open each of the ~11 one-level-deep `references/*.md` files (e.g. "Open `references/scopes-test.md` when writing **declarative** scope assertions", "Open `references/common-failures.md` when a test won't run…"). References are one level deep, consistently pathed, and the Related Skills section separates companion-skill loading from internal references — matching anchor 5's well-signaled, easy-to-navigate structure. | 5 / 5 |
Total | 19 / 20 Passed |