octocode-loop

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

• Third-party content exposure detected (high risk: 0.78). High indirect prompt-injection risk: the loop’s runtime tool calls (e.g., ghGetFileContent / ghSearchCode / npmSearch / web-surface reads) can ingest outsider-authored free text from remote repositories/web pages into the agent’s LLM context as “results”/file slices, which the loop then uses for subsequent actions.