mantis-patch

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

• Third-party content exposure detected (medium risk: 0.65). SKILL.md instructs the agent to “Load Reproduction Results” by reading JSON files from workspace/findings/ at runtime; those JSON entries are outsider-authored free text (e.g., findings produced by other parties/tools) that the agent then uses in its LLM-driven patch generation flow.