The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the use of orca terminal send and orca exec, which allow the agent to execute arbitrary commands within terminals managed by the Orca environment. This is a core functional requirement of the tool for developer workspace management.
[DATA_EXFILTRATION]: The skill provides access to sensitive data and network capabilities via orca cookie get (retrieving browser session cookies) and orca upload (transferring files). These features are standard components of the tool's integrated browser and workspace functionality.
[PROMPT_INJECTION]: The skill has a high attack surface for indirect prompt injection because it instructs the agent to ingest and act upon untrusted content from terminal outputs (orca terminal read) and browser state (orca snapshot).
Ingestion points: Untrusted data enters the agent context through orca terminal read, orca snapshot, and orca console commands defined in SKILL.md.
Boundary markers: The instructions do not define specific delimiters or warnings to ignore instructions found within ingested data.
Capability inventory: The agent can perform high-privilege actions based on ingested data, including orca terminal send (shell access), orca eval (JavaScript execution), and orca automations create (task scheduling).
Sanitization: There are no mentioned sanitization or validation steps for content processed from the terminal or browser tabs.
[REMOTE_CODE_EXECUTION]: The skill includes an orca eval command which permits the execution of arbitrary JavaScript within the context of the embedded browser tabs. This is an expected capability for a browser automation interface.

orca-cli