The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions suggest using npx -y @mvanhorn/printing-press-library to install the required CLI, which fetches code from the public npm registry.
[REMOTE_CODE_EXECUTION]: The installation process uses npx -y, which involves downloading and executing remote code without manual confirmation.
[DATA_EXFILTRATION]: The skill features a --deliver webhook:<url> flag that allows the output of any router command—potentially containing sensitive configuration data, IP addresses, and client lists—to be sent to an arbitrary external URL. It also supports a GL_INET_FEEDBACK_ENDPOINT which can be configured to automatically POST data to a remote server.
[COMMAND_EXECUTION]: The rpc call functionality allows for the execution of arbitrary RPC commands and UCI configuration changes on the router, providing a high-privilege control channel over the device.
[CREDENTIALS_UNSAFE]: The skill manages and stores router administrative credentials and SSH keys within credentials.toml and snapshot profiles.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its handling of untrusted environmental data.
Ingestion points: The skill reads external data from the router API, including WiFi SSIDs and connected client names.
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat router-sourced strings as untrusted data.
Capability inventory: The agent can invoke sensitive functions like rpc call, vpn toggle, and venue connect based on the processed data.
Sanitization: There is no evidence of filtering or sanitization of data ingested from the network environment before it is used by the natural-language command resolver (which).

pp-gl-inet