lookdev
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a human-in-the-loop workflow where an agent processes feedback (edits, highlights, comments) from an interactive studio. This creates a potential surface for indirect prompt injection if the user-provided feedback contains malicious instructions. \n
- Ingestion points: The agent ingests a machine-readable JSON patch from the generated studio (detailed in
SKILL.mdunder 'Text & media lookdev').\n - Boundary markers: The skill suggests validating the patch schema and checking bounded text sizes.\n
- Capability inventory: The agent has file read/write and command execution capabilities (via tools like
claude-code).\n - Sanitization: The instructions explicitly mandate treating pasted text/comments/metadata as untrusted data and ensuring they are never interpreted as executable HTML or instructions.\n- [SAFE]: The skill includes proactive security measures for its core functionality, such as instructing the agent to bind local servers only to loopback addresses, avoid directory traversal, and keep personal data out of URLs.
Audit Metadata