The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions explicitly direct the agent to bypass sandbox restrictions by re-running commands with escalated permissions if a runtime_access_denied error occurs. This encourages the agent to seek higher privilege levels to circumvent platform security controls.
[COMMAND_EXECUTION]: The skill relies on resolving and executing local binaries (orca, orca-ide, or orca-dev) based on the presence of environment variables like ORCA_CLI_COMMAND and ORCA_DEV_REPO_ROOT.
[INDIRECT_PROMPT_INJECTION]: The skill utilizes a discovery stub pattern to dynamically fetch its full instruction set from the output of a local command (ORCA skills get orca-cli). This creates a vulnerability where the agent's behavior is dictated by external tool output.
Ingestion points: The output of ORCA skills get orca-cli is loaded into the agent's context as a guide.
Boundary markers: None present; the skill does not use delimiters to isolate the dynamically loaded guide from its base instructions.
Capability inventory: The skill manages worktrees, terminals, repositories, and an embedded browser through terminal commands.
Sanitization: No sanitization or validation is performed on the output of the CLI command before it is used to guide agent behavior.

orca-cli