analyze-adversarial-report

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

• Third-party content exposure detected (high risk: 0.85). The skill ingests outsider-authored free text from Coval adversarial report artifacts at runtime—e.g., it fetches and reads transcripts via curl https://api.coval.dev/eval/transcript?... | jq -r '.data.transcript', where the transcript content includes the adversarial persona’s interactions and scenario text authored by others—so that readable text can be fed into the agent’s LLM context for analysis.