The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install external CLI tools using npx -y @mvanhorn/printing-press-library and go install github.com/mvanhorn/printing-press-library/.... While these resources originate from the skill's author, they represent the execution of third-party code.
[COMMAND_EXECUTION]: The skill implements a complex 'Automatic learning' system (teach, recall, playbook) that allows for dynamic command generation and execution. It uses 'playbooks' which contain executable steps stored in a local SQLite database and retrieved based on natural language queries. This creates a significant surface for local database poisoning where malicious 'learnings' could lead the agent to execute unintended system commands.
[DATA_EXFILTRATION]: The CLI tool supports a --deliver webhook:<url> flag, which explicitly allows the output of any command (including oven telemetry, camera frames, and local logs) to be transmitted to an arbitrary remote URL.
[CREDENTIALS_UNSAFE]: The skill manages sensitive authentication material (signing keys and access tokens) stored at ~/.config/juneoven-pp-cli/identity.json and credentials.toml. While the documentation recommends setting appropriate permissions (0600), the skill provides instructions for the agent to interact with these files.
[DYNAMIC_EXECUTION]: The 'playbook' system functions as a runtime script engine, substituting entity slots into stored command templates and executing them via the CLI. This dynamic assembly of shell commands is an inherent risk factor.

pp-juneoven