The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill explicitly instructs the agent to "re-run it with escalated permissions" if a runtime_access_denied error occurs. This encourages the agent to attempt to gain administrative or root access (e.g., via sudo) to bypass system sandboxing or security policies.\n- [COMMAND_EXECUTION]: The agent is directed to execute various system commands using the orca CLI tool. The specific binary path is determined by environment variables like ORCA_CLI_COMMAND, which could be manipulated to execute untrusted code if the environment is compromised.\n- [DYNAMIC_EXECUTION]: The skill is a "discovery stub" that dynamically loads its primary operating guide using the command ORCA skills get orca-cli. This architecture hides the true capabilities and instructions of the skill from static analysis, as they are fetched only when the skill is invoked.\n- [INDIRECT_PROMPT_INJECTION]: The dynamic loading of instructions creates a vulnerability surface for indirect prompt injection.\n
Ingestion points: The output of the command ORCA skills get orca-cli described in SKILL.md.\n
Boundary markers: None; the agent is instructed to follow the guide fetched from the command output.\n
Capability inventory: The skill description lists capabilities including operating terminals, worktrees, repositories, and an embedded browser.\n
Sanitization: No validation or sanitization is performed on the content returned by the CLI before the agent processes it as instructions.

orca-cli