The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies sandbox command output and remote documentation as untrusted ingestion points. It provides clear instructions for the agent to attribute these outputs and implement sanitization, such as using isolated staging directories and prohibiting the local execution of guest-generated code.\n- [DYNAMIC_EXECUTION]: The core functionality involves executing code within disposable Linux microVMs. The instructions focus on maintaining this isolation through proper resource lifecycle management and explicit cleanup procedures for both successful and failed tasks.\n- [EXTERNAL_DOWNLOADS]: The skill fetches up-to-date documentation from official vendor domains (render.com) and encourages the use of official SDKs. It includes a proactive security warning regarding the PyPI package registry to ensure developers install the correct, versioned package.\n- [CREDENTIALS_UNSAFE]: The skill implements best practices for credential safety by explicitly instructing the agent never to transfer API keys into the sandbox and to filter out sensitive files like .env or private keys during resource creation.

render-sandboxes