Swarm Coord

• Risk level: Low
• Analyzed: Apr 1, 2026, 6:59 PM
• Findings: 0
• Version: 1.0.0

Overview

The skill's description, instructions, and requirements are internally consistent for a multi‑agent orchestration helper, but it persists task state and can forward context (including file paths/config values) so you should review how that data and downstream agent permissions are handled before use.

This skill appears to do what it says: it decomposes a job, spawns sub‑agent sessions, monitors them, and aggregates results. Before installing or running it: 1) Confirm where '.task-state.json' will be written and how long it is retained; sensitive data should not be stored there. 2) Review what “上下文（文件路径、配置值等）” you allow the CEO to forward—avoid sending secrets or whole config files unless necessary. 3) Ensure child agents have only the minimal credentials/permissions needed (e.g., Git or messaging tokens) and are trusted to perform external actions. 4) If you use Memory‑Dream or other memory skills, decide whether you want swarm outputs persisted to long‑term memory. 5) Test the skill in a low‑privilege environment first (no access to production repos/secrets) to observe how it passes context and handles failures. If you need, request the author/sources for the optional install repo to verify there are no hidden scripts in the packaged version.