The Agent Skills Directory

[SAFE]: The skill acts as a security hardening utility and explicitly prohibits insecure configurations, such as the use of the privileged flag or mounting sensitive host directories. \n- [EXTERNAL_DOWNLOADS]: The skill references the official node:20-alpine image from Docker Hub, which is a trusted and well-known source for creating minimal, secure container environments. \n- [COMMAND_EXECUTION]: While providing templates for Docker run commands, the skill does not execute shell commands directly and requires explicit user confirmation before writing configuration files to disk. \n- [PROMPT_INJECTION]: The skill processes untrusted metadata from external skills to generate sandboxes. 1. Ingestion points: Skill metadata input section in SKILL.md. 2. Boundary markers: No explicit markers are present in the generation prompts. 3. Capability inventory: fileWrite permissions are limited to a dedicated sandbox output folder. 4. Sanitization: Safety rules 8 and 9 mandate restricted file paths and manual user review before any action is taken.

sandbox-guard