Content
62%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 highly actionable, well-sequenced workflow with strong validation checkpoints, but it pays for that with significant verbosity and a monolithic structure that inlines several large blocks that should live in reference files. Trimming the repeated enforcement language and extracting the persistence script and audit checklists into bundle files would improve both conciseness and progressive disclosure.
Suggestions
Extract the ~130-line persistence bash script (Step 8) into a scripts/ helper file and invoke it with arguments, cutting a large chunk of inline tokens.
Move the AI Surface Audit checklist (Step 7b) and the critique prompt template (Step 5b) into references/ files, keeping only a short pointer plus the output contract inline.
Collapse the repeated MANDATORY/HARD-GATE emphasis (the same requirement is stated in the frontmatter, the gate block, and step headers) into a single statement of the 3-search minimum.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~590-line body is noticeably verbose: repeated 'MUST'/'HARD-GATE' emphasis, emoji banner templates, and a ~130-line inline persistence bash script that belongs in a helper file. It also restates the same enforcement message multiple times (e.g., the search.py mandate appears in the frontmatter trigger, the HARD-GATE block, and Step 4's header). | 2 / 5 |
Actionability | Guidance is largely copy-paste executable: concrete search.py invocations with flags, an executable preflight check, a contrast checker with exit-code semantics, and full AskUserQuestion templates. Minor gaps remain — e.g., '$DESIGN_BODY' is validated for emptiness but the skill never shows how it is populated, and the Figma/MCP calls depend on unverified tool availability. | 4 / 5 |
Workflow Clarity | Steps 1-8 are explicitly sequenced with numbered phases, and validation is embedded throughout: preflight gating on READY, dial range validation before every search call, contrast-checker exit-1 fix-and-rerun loops, supersede-self guards, and explicit failure remediation for each missing-resource state. Feedback loops (fix palette and re-run) are present for the risky operations. | 5 / 5 |
Progressive Disclosure | The body has clear section structure (numbered steps, phase headers), but substantial content that belongs in separate reference files is inlined — the 130-line persistence script, the AI Surface Audit checklist, and the critique prompts. References to 'skills/blocks/design-taste.md' point outside this bundle (no references/ directory exists) and are not clearly signaled as navigable paths. | 3 / 5 |
Total | 14 / 20 Passed |