The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of shell commands and the launching of other agents (e.g., Claude, Codex) through orca terminal create --command and orca terminal send functionalities. It also allows running JavaScript within the browser via orca eval.
[DATA_EXFILTRATION]: The skill grants access to terminal output via orca terminal read and browser content via orca snapshot, creating a path for the agent to access and process potentially sensitive workspace data.
[PROMPT_INJECTION]: The skill handles untrusted data from browser sessions and terminal outputs, posing a risk of indirect prompt injection. The documentation explicitly advises the agent to treat fetched page content as untrusted data rather than instructions.
Ingestion points: Browser snapshots (orca snapshot), terminal history reads (orca terminal read), and mobile emulator data.
Boundary markers: The instructions include a specific textual warning to the agent to treat fetched content as data, not as instructions.
Capability inventory: Extensive command execution capabilities including terminal creation (orca terminal create), arbitrary browser script execution (orca eval), and the ability to schedule recurring tasks (orca automations create).
Sanitization: No explicit sanitization or input validation mechanisms for external content are described in the tool usage patterns.

orca-cli