atproto

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

• Third-party content exposure detected (high risk: 0.85). Yes: the required runtime workflow is to subscribe to a relay’s subscribeRepos firehose and process incoming #commit/#identity/#account event payloads (outsider-authored user content from other accounts) which are parsed into readable JSON/strings and then used by the agent’s LLM context.