The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of local shell commands (e.g., sage war, sage fleet, sage judge publish) to monitor environment state and update the coordination layer. These operations are core to the skill's stated purpose of providing a live advisory session for the 'sage' ecosystem.
[DATA_EXFILTRATION]: While the skill reads JSON data from the local environment and session metadata (via sage commands), all operations are local. There are no network requests, external data transfers, or connections to remote domains identified.
[PROMPT_INJECTION]: The skill defines a set of 'Hard rules' that limit the agent's behavior (e.g., forbidding it from editing product trees or claiming globs). These are behavioral constraints designed to ensure the agent maintains its role as a monitor rather than an active worker, and do not constitute malicious prompt injection.

sage-judge