Content
88%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 dense, highly actionable eight-step workflow with concrete MCP tool names, ref formats, a copy-paste shell command, and explicit dry-run validation with fix-and-retry feedback loops. The main weaknesses are the over-packed run-on paragraph in step 8 and the absence of any reference-file split for the skill itself, leaving detail inline that would fit better in a reference file.
Suggestions
Move the dry-run/temp-directory mechanics of workflow step 8 (server temp-dir restriction, inline content limit, copy-then-pass workaround, hollow-descriptor diagnosis) into a one-level-deep reference file such as references/dry-run.md, keeping a short 'dry-run every blueprint, fix the shipped file and retry' instruction in the main step.
Break the remaining step 8 text into sub-bullets or a short checklist so verification, failure handling, and cross-references are scannable instead of a single paragraph.
Move the generated-skill 'Reference Files' descriptions alongside the 'Generated Skill Layout' tree so the file listing and per-file contents are in one place.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes competence — no explanations of what MPS or DSLs are, and every tool name and gotcha is project-specific knowledge Claude would not have. Not 5: workflow step 8 is an over-packed run-on paragraph that mixes verification, temp-directory mechanics, inline-content limits, a shell snippet, failure policy, and a cross-reference, and could be tightened or split. | 4 / 5 |
Actionability | Quotes concrete tool calls ('Call `mps_mcp_get_project_structure` with `includeModels: false`', 'mps_mcp_insert_root_node_from_json with `dryRun: true`'), exact ref formats ('l:<uuid>:<languageName>'), a copy-paste shell snippet ('d="${TMPDIR:-/tmp}/<dsl-name>-blueprints"; mkdir -p "$d"; cp ...'), a layout tree, and the blueprint JSON shape ('properties:[{name,value}]'), not a map'). Not 4: the guidance is fully executable with placeholders only where a generator skill cannot avoid them. | 5 / 5 |
Workflow Clarity | The 8-step workflow (discover → scope → examples → concepts → sample → generate → cross-link → verify) is clearly sequenced with an explicit verification step that includes dry-run validation and a fix-and-retry feedback loop ('When a dry run fails, fix the shipped file, not only the inline text or the temp copy, and dry-run again from the fixed file'). Not 4: checkpoints are explicit, not merely present. | 5 / 5 |
Progressive Disclosure | Clear section headers (When to Use, Prerequisites, Workflow, Generated Skill Layout, Requirements, Reference Files) make the body easy to navigate, and the generated-skill reference layout is well specified. Not 5: the skill itself ships no reference files, and step 8's dry-run/temp-directory mechanics are inlined in the main workflow where a one-level-deep reference file would fit better; not 3 because structure and signaling are otherwise good. | 4 / 5 |
Total | 18 / 20 Passed |