Replicate Skill

Quick start

Goal: operate on Replicate resources directly: models, predictions, trainings, files, and deployments.

Reference notes: skills/replicate-skill/references/replicate-skill.md

Common workflows

• Account and discovery: get_account, search_models, list_models, get_models, get_models_readme
• Run inference: create_predictions or create_models_predictions
• Poll jobs: get_predictions, get_trainings
• Manage training: create_trainings, cancel_trainings
• Manage deployments/files: create_deployments, update_deployments, list_files, create_files

Recommended workflow

1. Inspect the model or version schema first.
2. For short jobs, use Prefer: wait when appropriate.
3. For long jobs, create then poll until terminal status.
4. Save or download outputs you need; hosted outputs may be removed later.

High-risk actions

Require explicit confirmation before:

• deleting models, versions, deployments, or files
• canceling active predictions/trainings on behalf of the user
• changing live deployment configuration

References

• Condensed notes: skills/replicate-skill/references/replicate-skill.md