Back to skill
Skillv1.0.0
VirusTotal security
Agent Swarm - 多智能体集群编排 · External malware reputation and Code Insight signals for this exact artifact hash.
Scanner verdict
SuspiciousApr 29, 2026, 5:02 AM
- Hash
- f28ea9fad0528b086743a9b90f2b695386b131e27cacc1d2883961c460b3ff4b
- Source
- palm
- Verdict
- suspicious
- Code Insight
- Type: OpenClaw Skill Name: lx-agent-swarm Version: 1.0.0 The skill bundle is suspicious due to multiple explicit prompt injection vulnerabilities and the broad, powerful capabilities granted to several sub-agents. The `SKILL.md` and `scripts/experience_logger.py` explicitly describe and implement a mechanism where user-controlled 'experiences' are injected directly into sub-agent prompts, creating a clear prompt injection vector. Additionally, `scripts/agent_manager.py` allows user-provided names/emojis to be embedded into agent persona files (`SOUL.md`, `AGENTS.md`), which are also read by agents, posing another injection risk. These vulnerabilities are critical because agents like `coder`, `analyst`, and `automator` are explicitly configured with `exec` and `process` tools (as seen in `references/chj-private-config-template.json` and `references/setup-guide.md`), enabling potential remote code execution if exploited by a malicious user. While the code itself doesn't exhibit intentional malicious behavior, it provides a highly vulnerable framework.
- External report
- View on VirusTotal