The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill instructions include installing a global package omniroute via npm install -g. The update --apply command indicates a mechanism for downloading and executing remote code to update the utility on the host system.
[PERSISTENCE_MECHANISMS]: The skill provides specific subcommands (autostart enable, autostart disable, autostart toggle) designed to manage the automatic execution of the tool on the host system, which could be used to maintain persistent access.
[DATA_EXPOSURE_AND_EXFILTRATION]: Multiple commands are designed to handle sensitive data. The setup command accepts --api-key and --password flags, and the env set command allows the agent to modify environment variables. Furthermore, the env show --json command could potentially expose system secrets to the agent's execution context.
[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface where untrusted data could influence system-altering commands.
Ingestion points: Parameters provided to omniroute env set, omniroute config set, and omniroute setup.
Boundary markers: None identified in the skill instructions.
Capability inventory: Subprocess calls via the omniroute CLI to modify system persistence, manage environment credentials, and install/update global software packages.
Sanitization: No validation or sanitization logic is specified for the input parameters.

cli-setup