Multi-Agent Dev Team

2-agent collaborative software development workflow for OpenClaw

This skill appears to do what it says: coordinate a PM and Dev agent to create projects. Before you use it, take these precautions: - Configure a dedicated workspace (cwd) so the agents only write in a directory you control. - Do not provide GitHub, Vercel, or other deployment credentials unless you trust the skill and are ready for it to push or deploy code. - Review generated code and commits before deploying to production — agents can introduce bugs or insecure dependencies. - If you need higher assurance, test the skill on a throwaway project and inspect the created files and commit history first.

Type: OpenClaw Skill Name: multi-agent-dev-team Version: 1.0.0 The skill is designed for legitimate software development, but both the PM and Dev agents are explicitly granted high-risk capabilities like `exec` (arbitrary command execution) and `Read/Write` (arbitrary file system access) in their SOUL.md files. The Dev agent is also instructed to `git push` code. While these capabilities are necessary for a development agent, they introduce a significant prompt injection vulnerability. A malicious user could craft a project request that, when processed by the PM and then passed to the Dev agent, could lead to unauthorized arbitrary code execution or data exfiltration from the host system. There is no evidence of intentional malicious behavior by the skill's author, but the inherent risk of these powerful, agent-controlled tools makes it suspicious.