The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (LOW): The skill references and executes the Docker image ghcr.io/cardano-scaling/hydra-node. While this is the official image for the protocol, it represents a remote dependency pulled at runtime.
[COMMAND_EXECUTION] (LOW): The scripts/hydra-node.sh wrapper script uses exec to pass arguments directly to the container or local binary. This allows the agent to control CLI parameters, which is the intended purpose but grants broad tool access.
[DATA_EXPOSURE] (LOW): Troubleshooting steps involve operations on sensitive file paths such as cardano.sk (private signing keys). The skill contains explicit instructions for the agent to avoid reading the actual key contents, which mitigates the risk of accidental exposure.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill is designed to process external data that could be controlled by an attacker if a node is compromised.
Ingestion points: Node logs (hydra-node.log) and API responses from curl commands to local or peer health endpoints.
Boundary markers: None; the agent processes raw grep/curl output directly.
Capability inventory: File system access, network requests via curl, and Docker command execution.
Sanitization: No sanitization or escaping is performed on log entries before the agent interprets them.

hydra-head-troubleshooter