The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of code within remote Aliyun Bailian Sandbox instances via the bl CLI and runtime APIs such as envd.
[COMMAND_EXECUTION]: Includes script templates and instructions for Bash, Python, and Node.js to perform instance connectivity checks and orchestrate data-plane interactions.
[EXTERNAL_DOWNLOADS]: Interacts with official *.aliyuncs.com (Alibaba Cloud) domains for sandbox management and file uploads, which are recognized as well-known service endpoints.
[DYNAMIC_EXECUTION]: Provides code templates for Python and Node.js in assets/instance-connections.md that are intended to be generated and executed by the agent to facilitate browser automation and health monitoring.
[INDIRECT_PROMPT_INJECTION]: The skill's ability to automate browsers and ingest remote files creates a vulnerability to indirect prompt injection from untrusted external content.
Ingestion points: External web content retrieved through browser automation, sandbox instance metadata, and file uploads via bl sandbox file upload.
Boundary markers: Operational instructions in assets/instance-connections.md specify capturing connection tokens in private directories with restricted permissions (umask 077).
Capability inventory: Full lifecycle control of sandbox resources, remote shell access (envd), and browser automation (CDP/Puppeteer).
Sanitization: Mandates explicit user confirmation (--yes) for high-risk operations such as instance or template deletion.

bailian-sandbox