The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends installing its core software by piping a remote script from https://pilotprotocol.network/install.sh directly into the system shell, a method that allows for unverified code execution.
[COMMAND_EXECUTION]: The skill features an 'app store' capable of installing and running local binary applications (such as SQL databases, Docker containers, and browser drivers) which are spawned as background processes.
[DATA_EXFILTRATION]: The core protocol facilitates encrypted peer-to-peer communication and file transfers (pilotctl send-file), providing a direct path for data to be moved out of the local environment to external nodes.
[PROMPT_INJECTION]: The skill ingests 'live external data' from a directory of over 430 external service agents. This untrusted data is stored in the local file system and subsequently processed by the agent.
Ingestion points: External agent responses are saved as JSON files in ~/.pilot/inbox/.
Boundary markers: There are no instructions for using delimiters or boundary markers to prevent the agent from obeying instructions embedded in the external JSON data.
Capability inventory: The skill has access to the Bash tool, file system read/write, and the ability to execute network commands through pilotctl.
Sanitization: The skill lacks any prescribed sanitization or validation of the data retrieved from the network before it is processed by the AI.

pilotctl