The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill explicitly instructs the agent to "re-run it with escalated permissions" if execution is blocked by a sandbox (indicated by a runtime_access_denied error). This is a direct instruction to attempt to bypass security boundaries or acquire higher system privileges.\n- [DYNAMIC_EXECUTION]: The skill uses the command ORCA skills get orca-cli to load its operational guide dynamically from the local binary at runtime. This means the actual instructions the agent follows are determined at execution time, making them unverifiable through static analysis of the skill file.\n- [COMMAND_EXECUTION]: The primary function of the skill involves executing the orca CLI tool via shell commands to manage worktrees, terminals, and browsers.\n- [INDIRECT_PROMPT_INJECTION]: The skill description mentions "handoff" / "handover" functionality, which creates a channel for one agent to share state and data with another. This introduces a surface for indirect instructions to be passed between agents.\n
Ingestion points: Data and state shared during "handoff" operations via Orca-managed worktrees and terminals.\n
Boundary markers: No instructions are provided to the agent to use delimiters or ignore instructions embedded in the shared state.\n
Capability inventory: The CLI provides powerful capabilities including shell access, file management, and embedded browser control.\n
Sanitization: No sanitization or validation of the data being handed over between agents is defined.

orca-cli