together-core-workflow-a
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill follows secure operational procedures for interacting with third-party AI training services.
- [EXTERNAL_DOWNLOADS]: The skill and its references point to official Together AI resources, including documentation at
docs.together.aiand source code atgithub.com/togethercomputer. These are recognized well-known services for the AI industry, and the references are provided neutrally for configuration and SDK usage. - [CREDENTIALS_SAFE]: The instructions explicitly mention using project-scoped credentials (
TOGETHER_API_KEY) and emphasize that keys must remain in an approved secret store and be redacted in manifested output. - [INDIRECT_PROMPT_INJECTION]: The skill involves processing training datasets which could contain untrusted data. However, it mitigates this risk through a mandatory evidence chain:
- Ingestion points: The skill ingests a
training-filepath provided as an argument. - Boundary markers: Explicit instructions require human cost approval and dataset authority confirmation before any upload or job submission occurs.
- Capability inventory: The skill uses standard file system tools (
Read,Glob,Grep,Write,Edit) and network tools (WebFetch) to interact with the Together AI API via the Python SDK. - Sanitization: Instruction step 2 mandates validation of format, licenses, and data leakage before processing.
- Ingestion points: The skill ingests a
Audit Metadata