The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The instructions explicitly guide the agent to re-run commands with escalated permissions (such as using sudo or administrative accounts) if it receives a runtime_access_denied error. This is a direct attempt to bypass security boundaries and sandbox constraints.
[INDIRECT_PROMPT_INJECTION]: The skill uses a mechanism to fetch operational guides dynamically via the command ORCA skills get orca-cli. This creates a vulnerability surface where the content returned by the CLI influences the agent's instructions at runtime.
Ingestion points: The agent processes output from the ORCA skills get orca-cli command as instructions (SKILL.md).
Boundary markers: There are no visible delimiters or instructions provided to the agent to ignore potentially malicious content within the dynamically loaded guide.
Capability inventory: The skill enables broad capabilities including terminal interaction, filesystem access, and repository management via the orca CLI.
Sanitization: The instructions do not specify any validation or sanitization of the content retrieved from the CLI before it is adopted into the agent's context.

orca-cli