The Agent Skills Directory

Remote Command Execution: The skill allows for the execution of shell commands within a remote sandbox environment via an HTTP API.
Context: This is a core functionality intended for validating SDK changes and troubleshooting issues in an isolated container.
Evidence: Use of HTTP POST requests to the /exec endpoint with shell-wrapped command payloads.
External Network Interaction: It performs network requests to an external service URL provided through host-injected environment variables.
Context: These requests are used for sandbox lifecycle management and retrieving configuration data like OpenAPI specifications.
Evidence: Usage of curl targeting the $SANDBOX_WORKER_URL throughout the skill documentation.
Command-Line Tool Usage: The skill utilizes standard CLI utilities to interact with the sandbox API and process its responses.
Context: Tools like curl, jq, base64, and awk are used to manage communication and decode stream data.
Evidence: Instructions for piping API output into jq for parsing and base64 for decoding stdout/stderr chunks.
Credential Security: The skill provides guidance on protecting sensitive authorization tokens.
Context: It instructs the agent to use bearer tokens in headers rather than query parameters to prevent exposure in logs and history.
Evidence: Explicit warning in the Credentials section of the SKILL.md file.

sandbox-bridge