lora-training
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [SAFE]: The skill contains explicit instructions to maintain security hygiene, such as using integrated authentication flows instead of requesting tokens in chat and ensuring that signed URLs and credentials are never exposed in user-facing text. \n- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it ingests and processes external data. \n
- Ingestion points: Media files are fetched via
upload_dataset_file_from_urland caption text is written viaupload_dataset_text_file(SKILL.md). \n - Boundary markers: The instructions use platform-defined paths but do not specify unique delimiters or explicit instructions to ignore potentially malicious content within the ingested files. \n
- Capability inventory: The agent can perform significant actions including submitting training jobs (
submit_training_job), managing job states, and checking account balances (get_balance). \n - Sanitization: There are no instructions for sanitizing or validating the content of the media or text files before they are processed by the training tools. \n- [EXTERNAL_DOWNLOADS]: The skill facilitates the download of media assets from user-provided URLs using the
upload_dataset_file_from_urltool to prepare datasets for AI training.
Audit Metadata