The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill explicitly instructs the agent to bypass sandbox security blocks by attempting to re-run commands with elevated privileges.
Evidence: "If it fails with runtime_access_denied, your sandbox blocked the connection: re-run it with escalated permissions".
[REMOTE_CODE_EXECUTION]: The skill functions as a discovery stub that dynamically fetches and executes additional instructions from a local command-line tool, a form of remote instruction loading.
Evidence: ORCA skills get orca-cli is used to "load the version-matched guide".
[COMMAND_EXECUTION]: The skill relies on executing various shell binaries (orca, orca-ide, orca-dev) and suggests resolving these paths based on environment variables or trial-and-error.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection because it loads instructions from the output of an external command without sanitization or clear boundary markers.
Ingestion points: The output of ORCA skills get orca-cli is treated as the primary instruction set for the session.
Boundary markers: No delimiters or safety instructions are provided to the agent to treat the fetched content as data rather than instructions.
Capability inventory: The skill possesses the ability to execute shell commands and instructions to request permission escalation.
Sanitization: No sanitization or validation of the fetched instruction content is performed.

orca-cli