Clawflow

• Analyzed: May 12, 2026, 4:17 AM
• Findings: 1
• Version: 1.0.0
• Hash: 8695cb3224b672869f3d27370d2bce80913b720d1c0233c3898ba2afcf211543
• Source: palm
• External report: View on VirusTotal

Findings (1)

The OpenClaw AgentSkills skill bundle is classified as suspicious due to a significant prompt injection vulnerability inherent in its design. The `SKILL.md`, `agent-loop.md`, `coordinating.md`, and `schemas.md` documents explicitly instruct the AI agent to embed 'upstream results' directly into the 'Context' section of dispatch messages for dependent subtasks. This means that if a malicious or compromised sub-agent returns crafted instructions or payloads as its 'results', the coordinating agent is designed to forward these unsanitized instructions as context to subsequent agents, potentially leading to a chain of prompt injections and unauthorized control over other agents in the DAG. The `init.py` and `message.py` scripts are utility files that appear benign and do not introduce direct vulnerabilities or malicious behavior themselves.