Multi Agent

A multi-agent collaboration system coordinating specialized AI roles—planner, executor, reviewer, and coordinator—to analyze, execute, and review complex tas...

What to consider before installing/running: - Missing/deferred dependency: The code requires a sibling package (../../react-agent) but package.json doesn't declare it. npm install may not fetch that dependency — verify you have the required react-agent package or update package.json before running. - Local file & command access: The Executor is designed to call tools that can read/write files and run shell commands. Only run this skill in a safe, non-sensitive environment (e.g., an isolated VM or container) until you audit the tools and ReActEngine it uses. - Command injection risk: The Executor builds a shell command by interpolating the task string into exec without escaping. If you will pass untrusted or user-supplied tasks, this is dangerous. Consider patching the code to use execFile/spawn with argument arrays or properly escape inputs. - Audit external code: Inspect the ToolsRegistry and ReActEngine implementations (the referenced react-agent package). Those components will determine what external endpoints or system resources the skill can access. - Run tests in sandbox: Use the included tests/examples in an isolated environment to confirm behavior before using on real data. Prefer to run with reduced toolsets (no tools that access network or sensitive files) while you evaluate. - If you need help: Ask the skill author to (1) list required dependencies in package.json, (2) document exactly which built-in tools the Executor will enable, and (3) fix the unsafe exec usage or document input sanitization requirements. Given these inconsistencies and the unsafe shell usage, treat this skill as suspicious until the above issues are resolved.

Type: OpenClaw Skill Name: yuyonghao-multi-agent Version: 0.1.2 The skill bundle implements a multi-agent orchestration framework with roles for planning, execution, and review. It is classified as suspicious due to a significant shell injection vulnerability in `src/agent-roles.js`. Specifically, the `ExecutorAgent` class contains a fallback execution path that uses `child_process.exec` to run an `echo` command with unsanitized string interpolation of the `task` input, which could allow for arbitrary command execution. While this represents a high-risk vulnerability (RCE), there is no evidence of intentional malice, data exfiltration, or hidden backdoors within the provided logic or documentation.