github

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 when the agent uses Apideck Issue Tracking endpoints (e.g., listing/reading GitHub issues and comments) because those responses include body text authored by GitHub users other than the operating user, which the agent then passes into the LLM; the SKILL.md also describes an escape-hatch Proxy that can fetch arbitrary GitHub endpoints returning outsider-authored prose.