stitch-extract-design-md
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code from frontend projects, which creates a potential surface for indirect prompt injection if the source files contain malicious instructions.
- Ingestion points: Reads project files including
package.json, CSS/SCSS/Less stylesheets, and framework-specific component files (React, Vue, Svelte, Angular) as described inSKILL.mdandreferences/workflow.md. - Boundary markers: The instructions lack specific structural delimiters for untrusted data in the generated output, though they mandate distinguishing between "observed facts" and "suggestions."
- Capability inventory: The skill performs file system read operations on the source tree and file system write operations to create
.stitch/DESIGN.md. - Sanitization:
SKILL.mdincludes explicit instructions to replace personally identifiable information (PII) with demonstration values and prohibits the inclusion of keys, secrets, or session tokens in the output.
Audit Metadata