Content
82%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.
A tightly written operational skill: every step is backed by runnable code with encoding details, an edge-case re-query, and a concrete output template, and the organization is easy to navigate. The residual improvements are minor — trimming the illustrative private-set list that duplicates the stated rule, and adding a final spot-check of the emitted URLs.
Suggestions
Trim Step 3's list of known private sets to two or three examples — the stated rule "if name.startswith('.')", skip it" already carries the instruction, and the full inventory is rediscovered on every run anyway.
Add a final verification step to the workflow (e.g., spot-check one or two generated URLs via the /nodes endpoint or confirm no CANVAS node-IDs slipped into the output) to close the loop on Step 4.
Consider moving the default file key/URL and the CDS-specific inventory data into a small reference file so the SKILL.md body stays a lean procedure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and operational — no explanation of what Figma or REST APIs are, code blocks carry the weight, and edge cases are stated in one or two sentences. Not a 5 because the 11-line list of known private sets in Step 3 ("Carousel: .count, .carouselPagination... Select: .Input Chip...") duplicates what the stated rule "if name.startswith('.'), skip it" already conveys and could be trimmed to a few examples. | 4 / 5 |
Actionability | Fully executable copy-paste-ready curl commands with inline python3 parsers for both the file and /nodes endpoints, including URL-encoding notes ("URL-encode : → %3A and , → %2C") and a diagnostic fallback ("print child types to diagnose"), plus a concrete URL template and a rendered output example. Common cases and the SECTION edge case are both covered. | 5 / 5 |
Workflow Clarity | A clearly sequenced 4-step process with a diagnostic fallback and an explicit error-recovery path ("When depth=2 returns only SECTION entries... re-query those section IDs"), which serves as a feedback loop. Not a 5 because there is no verification of the final output (e.g., spot-checking a generated URL or node type after formatting) — Step 4 mentions checking node types only "if you're unsure", leaving the last checkpoint implicit. | 4 / 5 |
Progressive Disclosure | Well-organized single-file skill with clear section headers (Overview, Steps 1–4, Output format, Tips), decision tables, and no references at all — nothing is buried or nested. Not a 5 because at ~160 lines the file-specific data (the known private-set inventory, the default file key/URL) sits inline where a one-level-deep reference file could keep the core procedure tighter, per the anchor structure. | 4 / 5 |
Total | 17 / 20 Passed |