Stitch Code-to-Design

The skill explicitly reads frontend code, writes .stitch/DESIGN.md, and uploads HTML and a design system; the data flow is partly visible, but the allowed Bash, Write, web_fetch, and Stitch tool scope is broad. It does not require per-upload confirmation, explain sensitive-data handling or API-key protection, or provide rollback/revocation. Its instruction not to block on verification further reduces user review, so points are deducted. No malware, credential theft, covert exfiltration, or destructive default is evidenced.

The sequence and prerequisites are reasonably clear, but the target file does not include the dependent skills, scripts, parameter schemas, or abnormal-input handling. Step 3 refers to extract-design-md while Step 4 invokes manage-design-system, leaving dependency interpretation to the user. There is no target-skill test coverage, failure feedback, retry strategy, or partial-completion recovery guidance, so only limited static plausibility is credited.

The audience, inputs (a built directory, index.html, assets, and projectId), and main outputs are fairly clear. However, the “always use” trigger for save, migrate, or upload is overly broad, and non-fit cases, framework boundaries, dynamic-app limitations, and Chinese-language behavior are unspecified. Core operation depends on Stitch MCP, an API key, and an external service; reachability from mainland-China networks is not evidenced, so points are deducted.

The target file has a readable prerequisite/workflow structure and links to dependent skills; the repository README documents installation, and repository context supplies an Apache-2.0 license and security policy. The skill itself lacks versioning, changelog, maintenance ownership, troubleshooting, stable parameter documentation, and complete examples. Dependency implementation details are absent from the supplied evidence, preventing a higher score.

The chain covers static HTML extraction, design-system extraction, and upload, so it plausibly addresses the core code-to-design task. However, there is no skill-specific reusable output example or execution evidence; verification is optional and non-blocking, leaving upload correctness, asset completeness, and post-failure usability for manual checking. Under the static calibration ceiling, this earns only a moderate-low score.

The target file supplies auditable workflow steps, paths, and parameter requirements, and the README gives positioning and usage examples. The supplied CI only validates a react-components example and does not cover code-to-design paths. There are no committed tests for this skill, third-party reproduction records, or result evidence, so the conclusion rests mainly on one static document.