stackblitz-core-workflow-a
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides defensive guidelines for developing browser-based IDEs, explicitly recommending the isolation of untrusted content and the protection of host secrets.
- [EXTERNAL_DOWNLOADS]: The skill references official documentation from WebContainers (
webcontainers.io) for API references and integration guides. These references point to well-known service domains and do not involve untrusted remote code execution. - [INDIRECT_PROMPT_INJECTION]: The skill describes an architecture that handles untrusted data (project files and preview content) which could be a surface for indirect injection. It includes a mandatory security framework to mitigate this:
- Ingestion points: The architected system ingests user-controlled project files and external dependencies (SKILL.md).
- Boundary markers: The skill instructs developers to establish explicit trust-boundary maps and isolation for preview content (SKILL.md).
- Capability inventory: The system handles filesystem operations, terminal processes, and runtime events (references/official-docs.md).
- Sanitization: The skill mandates path validation, normalization of filesystem operations, and treating preview URLs as untrusted application content rather than trusted host UI (SKILL.md).
Audit Metadata