The Agent Skills Directory

[SAFE]: The skill instructions and tool definitions do not contain any malicious patterns or safety-bypass attempts. The workflow is well-documented and prioritizes user confirmation of task plans.
[DATA_EXFILTRATION]: Network communication is restricted to a local MCP endpoint (http://localhost:18801/mcp). No unauthorized external domains or data exfiltration patterns were detected.
[COMMAND_EXECUTION]: While the skill is designed to manage agents that perform coding tasks in isolated git worktrees, the orchestration logic itself uses structured tools and does not involve arbitrary or hidden shell command execution within the skill's own context.

claude-devfleet