replicate
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides interfaces for interacting with various machine learning models (LLMs, image generators) that process natural language prompts. This creates a surface for indirect prompt injection if an agent populates these prompts with untrusted data from external sources without validation.\n
- Ingestion points: Prompt inputs in
run_model.py,run_llm.py, andpredict.py.\n - Boundary markers: None present in the provided documentation examples.\n
- Capability inventory: SDK calls to Replicate's model hosting API.\n
- Sanitization: Not demonstrated in the basic usage examples.\n- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install standard libraries (
replicate,cog,torch,transformers) from official, well-known package registries.\n- [CREDENTIALS_UNSAFE]: The skill demonstrates secure credential handling by advising the use of environment variables for the Replicate API token and providing a clear placeholder (r8_xxxxxxxxxxxx) in its examples.
Audit Metadata