The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill is a comprehensive guide for the replicate CLI tool. It documents standard operations like running predictions, managing deployments, and fine-tuning models without introducing malicious or unauthorized commands.
[CREDENTIALS_UNSAFE] (LOW): The skill references the REPLICATE_API_TOKEN and provides an example of setting it. It uses a safe placeholder (<token-from-replicate.com/account>) and correctly suggests interactive authentication (replicate auth login) as an alternative.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill demonstrates output chaining, where model results are piped into other commands using {{.output[0]}}.
Ingestion points: Model outputs are directly ingested as inputs for subsequent model runs.
Boundary markers: None present; data is passed directly through the CLI.
Capability inventory: The skill enables full model execution and file-saving capabilities via the replicate binary.
Sanitization: No sanitization of model output is mentioned, which is standard for CLI tools but identifies a surface where untrusted model output could influence subsequent parameters.
[DYNAMIC_EXECUTION] (LOW): The documentation includes the replicate scaffold command, which generates local project structures for Node.js or Python. This is a standard developer-enablement feature for creating local environments from existing AI predictions.

replicate-cli