stitch-design-spec
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external data such as PRDs, functional design packages, and UI structure documents to generate its outputs (Workflow Step 1 in SKILL.md). This creates an attack surface for indirect prompt injection if source documents contain malicious instructions. The skill mitigates this by enforcing a strict 'trisection' prompt format ([Context], [Layout], [Components]) and requiring explicit labeling of demonstration data, maintaining clear boundaries between agent instructions and untrusted data.
- [COMMAND_EXECUTION]: The skill includes a Python script (
scripts/validate_package.py) for offline structural validation. The documentation (references/validation.md) instructs users to execute this script locally. Analysis of the script confirms it uses standard libraries and implements path traversal protections (is_relative_to(root)) when reading files, and does not perform any network operations or unauthorized file access. - [EXTERNAL_DOWNLOADS]: The documentation references a specific commit hash from a Google Labs repository (
google-labs-code/stitch-skills) for keyword references and prompt enhancement patterns. This is a reference to a well-known organization and does not involve automated runtime execution of remote code.
Audit Metadata