design
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes previously published artifacts which may contain malicious instructions from other users.
- Ingestion points: The skill uses the Artifact tool or WebFetch to read content from existing canvas URLs.
- Boundary markers: The instructions include clear warnings to the agent to treat extracted content as data only and never as instructions.
- Capability inventory: The agent can execute a local Node.js script to process files and use the Artifact tool to publish content.
- Sanitization: The
seed-canvas.mjsscript performs validation on filenames, titles, and manifest structure, and escapes HTML characters when embedding content.
Audit Metadata