Content
66%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 design consultation workflow itself is excellent: phased, concrete, and validated with feedback loops and fallbacks. But the skill buries it in hundreds of lines of generic framework boilerplate, and everything lives in one monolithic SKILL.md with no reference files, which costs token efficiency and navigability.
Suggestions
Move the gstack framework boilerplate (preamble bash, Voice, AskUserQuestion format, Completeness Principle, Contributor Mode, Completion Status Protocol, Telemetry, Plan Status Footer) into a shared framework file or a preamble reference that this SKILL.md points to, cutting the body roughly in half.
Extract the design knowledge catalogs (aesthetic directions, decoration/layout/color/motion approaches, font recommendations, blacklist and overused-font lists, AI slop anti-patterns) into a references/ file (e.g., DESIGN_KNOWLEDGE.md) linked from Phase 3, keeping only the decision framework inline.
Remove time-sensitive details from the main flow — e.g., 'best websites 2025' search strings and the pinned BUN_VERSION=1.3.10 — or relocate them to a setup/deprecated section so they don't go stale in the core instructions.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Roughly the first 340 lines are generic gstack framework boilerplate (session/telemetry bash preamble, 'Voice' persona, AskUserQuestion format, Completeness Principle, Contributor Mode, Plan Status Footer) unrelated to the design task, and the design phases add further padding such as 'best websites 2025' search strings and a pinned BUN_VERSION=1.3.10. This is noticeably-to-severely verbose; it is not a 1 because the Phase 0-6 content itself is dense and directive rather than explaining things Claude already knows. | 2 / 5 |
Actionability | Guidance is copy-paste ready throughout: exact bash setup checks, full $D command templates (variants/check/compare/iterate/extract), a concrete polling loop with the feedback JSON shape, a complete DESIGN.md template, and a 9-point preview-page spec. Specific examples cover the common paths (Path A AI mockups, Path B HTML fallback, plan vs. implementation mode). | 5 / 5 |
Workflow Clarity | Phases 0-6 are clearly sequenced with explicit validation checkpoints and feedback loops: a $D check quality gate per variant, file-polling feedback with a 10-minute timeout and AskUserQuestion fallback, a 'Here's what I understood... Is this right?' verification step, coherence re-checks after overrides, and a final confirm listing any agent-default decisions. Fallbacks for missing binaries (browse, designer) are spelled out at each step. | 5 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), and this ~890-line monolith inlines content that clearly belongs in separate files: the framework preamble/voice/telemetry sections and the font catalogs, aesthetic-direction lists, and blacklist tables belong in shared or reference files. Structure within the file is decent, but the inlining problem is severe, matching the 'content that clearly belongs in separate files is inlined' anchor rather than the merely 'could be better organized' one. | 2 / 5 |
Total | 14 / 20 Passed |