The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads official libraries including replicate and cog from standard package registries to facilitate API interactions.
[COMMAND_EXECUTION]: Uses the cog CLI tool to run local predictions and deploy machine learning models to the Replicate cloud environment.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it passes user-supplied input directly to external machine learning models without sanitization.
Ingestion points: The prompt and negative_prompt parameters in run_model.py, run_llm.py, and predict.py files.
Boundary markers: Absent; user input is not wrapped in delimiters to prevent instruction overriding.
Capability inventory: Network communication via the replicate library to perform remote model inference.
Sanitization: No evidence of input filtering or character escaping to mitigate malicious prompt content.

replicate