The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes arbitrary shell commands on a remote host (pi@raspberrypi.local) using ssh. It also relies on the just task runner to perform both local and remote operations, such as managing Docker containers and querying system state.
[CREDENTIALS_UNSAFE]: The skill includes instructions to programmatically extract secrets, including API keys and authentication tokens, from a .env file on the remote Raspberry Pi using grep commands. While it advises against displaying the entire file, it facilitates the exposure of individual credentials to the agent's context.
[EXTERNAL_DOWNLOADS]: The workflow involves downloading Docker images from the GitHub Container Registry (ghcr.io). These are legitimate vendor resources used for deployment.
[PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by ingesting data from potentially untrusted external sources.
Ingestion points: Processes output from remote Docker container logs (just bot-logs), GitHub Actions workflow status updates (gh run watch), and external health check JSON responses (just bot-health).
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore embedded instructions within these data streams.
Capability inventory: The skill possesses significant capabilities, including remote command execution (ssh), local command execution (just), and repository interaction (gh, git).
Sanitization: No sanitization, filtering, or validation is performed on the data ingested from logs or network endpoints before it is processed by the agent.

deploy