write-swarm-client

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

• Third-party content exposure detected (high risk: 0.80). The skill's training workflow explicitly reads tasks from a (potentially public/user-contributed) Hugging Face dataset via RouterTaskReader (see "dataset = RouterTaskReader(... huggingface_dat_repo = HuggingfaceDatRepo(...))" and the subsequent loop over dataset.generate_training_tasks()), which the agent ingests and uses to drive agent behavior and rewards, exposing it to untrusted third-party content that could carry indirect prompt-injection.