The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the uv package manager installer from the well-known astral.sh domain to handle dependency management within the automation environment.
[EXTERNAL_DOWNLOADS]: Retrieves plugins and extensions from GitHub repositories to provide extended capabilities to the AI agents.
[REMOTE_CODE_EXECUTION]: Executes user-defined scripts and entrypoints inside isolated sandboxes, which is the core intended functionality for running scheduled automations.
[COMMAND_EXECUTION]: Uses standard utilities such as curl and tar to interact with the OpenHands Cloud API and to prepare code packages for deployment.
[INDIRECT_PROMPT_INJECTION]: Processes natural language instructions as prompts for automations which have access to sandbox tools and stored secrets. While this represents a potential attack surface for untrusted data, the risk is inherent to the automation use case and is mitigated by the platform's sandbox isolation and the requirement for authenticated access.
Ingestion points: The prompt field in the /preset/prompt and /preset/plugin API requests (SKILL.md).
Boundary markers: None explicitly defined in the API request structure for separating instructions from data.
Capability inventory: Automations have full tool access including bash execution, file system manipulation, and network operations (SKILL.md).
Sanitization: Validation is performed on input lengths and cron expressions, but natural language prompts are passed directly to the agent (references/custom-automation.md).

automation