What problem does it solve?
This Skill prevents unsafe or inconsistent execution of Agent Skills by generating a controlled OpenSandbox/Docker environment from an existing skill directory and its declared permission risks.
Core Features & Use Cases
- Scans Skill permissions and risk signals: analyzes the target skill’s SKILL.md and its Python/Shell/markdown files to identify potential high-risk behaviors (e.g., secret access, deletion, external sending).
- Generates sandbox assets on demand: creates per-skill sandbox configuration, including sandbox.json, PERMISSIONS.md, a requirements.txt stub based on detected imports, and a sandbox/Dockerfile template.
- Performs safety-oriented security checks: produces warnings or blocks guidance based on detected CRITICAL/HIGH/MEDIUM patterns, aligning the generated permission level to the observed needs.
- Updates agent documentation for operational compliance: ensures the project’s AGENTS.md includes a standardized “Sandbox 执行要求” section when missing.
- Builds a dedicated Docker image: constructs a
{skill-name}-sandbox:latest image from the generated sandbox directory and prepares a test run command.
Example use case: You have a community or internal skill that needs to run inside OpenSandbox; use this Skill to generate a least-privilege sandbox setup that denies dangerous filesystem paths, restricts environment variables, and blocks unexpected network/exfiltration behaviors.
Quick Start
Ask the agent to run the sandbox configurator for your skill named contract-legal-review, ensuring it scans the skill, generates sandbox.json and related files, performs security checks, and builds the Docker image.
