openclaw-guide

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). Outsider free text can enter the LLM context via inbound messages from external users on runtime channels (e.g., Telegram/Discord/Slack), which the gateway routes into the agent’s session for response generation.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The SKILL.md includes a direct install command that fetches and executes remote code (curl -fsSL https://openclaw.ai/install.sh | bash), which is an installation-time runtime dependency required for the skill to function and therefore executes remote code.