Content
85%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 well-sequenced, self-contained workflow with concrete rules, an explicit priority order, and genuine error-recovery feedback loops, and it correctly delegates parameter mechanics instead of duplicating them. It loses points only on mild emphasis redundancy and the lack of a worked example of a get_design_context call or its adapted output.
Suggestions
Trim the duplicated mandate structure — state each rule once instead of a 'You MUST X' sentence followed by a restating elaboration (e.g., merge the two sentences in steps 1–3).
Add one short worked example: a minimal get_design_context call on a node URL and a snippet of its return being adapted to a project component, covering the most common path end-to-end.
Move the skillNames logging-parameter paragraph into a brief note at the end (or the workflow's step 1) — it is invocation bookkeeping, not workflow guidance, and interrupts the direction/scope framing.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Quote: 'It is your primary tool — a single call returns reference code, a screenshot, and contextual hints' — information Claude cannot know elsewhere, so most tokens earn their place; but repeated emphasis ('You MUST' appears ~12 times, 'must NOT be skipped' style framing, 'This is a logging parameter used to track skill usage — it does not affect execution') is minor padding that could be trimmed, matching anchor 4. Not 5 because the duplicated mandate structure restates each rule twice (the MUST sentence plus the elaboration); not 3 because there is no over-explanation of concepts Claude already knows. | 4 / 5 |
Actionability | Quote: 'never hand-write or inline <svg>/<path>', 'a fixed-size container (icons are usually square, e.g. size-[24px], overflow-clip) with BOTH width and height set', 'expires in ~7 days — download-and-commit the exact asset bytes' — concrete, executable guidance with specific class names and a 5-level hint priority; as an instruction-only skill the absence of code is not penalized, but a couple of gaps remain (no example get_design_context call or adapted-output pattern; 'wire a dynamic content image to the project's data source' is high-level), matching anchor 4. Not 5 because the common cases (a sample call and one adapted component example) are not shown; not 3 because the guidance is directly executable, not pseudocode. | 4 / 5 |
Workflow Clarity | Quote: a numbered 5-step Workflow ('Call get_design_context first' → 'Treat the output as a reference' → 'Reuse what the project already has' → 'Honor the response hints by priority' → 'Reproduce images and icons faithfully') plus an explicit Error Recovery section ('STOP and read the message before retrying', 'retry against a smaller node', 'ask the user for a node-specific URL'), matching anchor 5's clear sequence with feedback loops for error recovery. Not 4 because error-recovery checkpoints are explicit and the hint priority order is unambiguous; no destructive/batch operations exist, so the validation cap does not apply. | 5 / 5 |
Progressive Disclosure | The ~55-line body is well-organized into Direction and Scope, Workflow, and Error Recovery, needs no bundle files (none exist), and its single reference — 'use [figma-generate-design](../figma-generate-design/SKILL.md) instead' — is one level deep and clearly signaled for direction disambiguation; parameter mechanics are correctly delegated to the tool description ('live on the get_design_context tool description itself — follow them there'), matching the simple-skill exception (well-organized sections, no external references needed) at anchor 5. | 5 / 5 |
Total | 18 / 20 Passed |