The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches source code and configuration from the vendor's official GitHub repository (different-ai/openwork).\n- [COMMAND_EXECUTION]: Utilizes the daytona CLI to create sandboxes and execute setup commands, such as pnpm install, git clone, and service initialization scripts inside the ephemeral environment.\n- [PROMPT_INJECTION]: The skill processes instruction-like steps from files within the repository (e.g., evals/daytona-flows.md). This constitutes an indirect prompt injection surface as the agent interprets and executes steps defined in external data.\n
Ingestion points: Markdown files in the evals/ directory within the cloned repository.\n
Boundary markers: Absent; there are no delimiters or instructions to prevent the agent from following malicious commands if they were to be present in the evaluation files.\n
Capability inventory: The agent has access to daytona exec for shell commands and browser_evaluate for executing JavaScript within the sandbox's browser.\n
Sanitization: The skill does not perform validation or sanitization of the steps read from the evaluation files before the agent processes them.

run-evals