mantis-calibrate

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

• Third-party content exposure detected (high risk: 0.85). The required workflow explicitly reads free-text/JSON content from workspace/findings/ (outsider-authored security findings and empirical results) and also reads workspace/kb/THREAT_MODEL.md, which can contain attacker-controlled text, then uses that content to generate LLM reasoning and outputs—so outsider free text can enter the LLM context.