sandbox-guard
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The generated Dockerfiles reference the
node:20-alpineimage from Docker Hub, which is a well-known and standard base image.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided skill names and permission sets to generate configuration files.\n - Ingestion points: Input placeholders like
<name>and permission lists are used to populate Dockerfiles and shell commands.\n - Boundary markers: The skill does not define specific delimiters for these inputs to prevent potential code or command injection into the generated output.\n
- Capability inventory: The skill has
file-writeaccess to save results to the.openclaw/sandbox/directory.\n - Sanitization: No specific sanitization or escaping of the user-provided input is defined before its inclusion in the generated output. However, the skill mandates a human review step (Rule 9) before any files are written to disk, which significantly mitigates this risk.
Audit Metadata