amplifier-tool-leverage-patterns

To decide between a Python library and a CLI wrapper, assess your consumers: use libraries for direct imports, and CLI wrappers for shell-out execution. Keep logic in one home and expose thin adapters to avoid duplication.

The minimal packaging surface for an agent-callable workflow module is the public-interface isolation layer. It exposes only the thin adapter needed for agent invocation while keeping deterministic logic in a single shared home.

Use a dot pipeline instead of a Python library when you need Resolve composition without exposing importable APIs. Pipelines work for declarative chaining, while libraries suit consumers requiring direct programmatic integration.

To keep logic DRY across a CLI, Python API, and agent tools, isolate a single source of truth for your business logic. Expose only thin adapters for each consumer type, preventing duplicated implementation across repos.

Yes, you need to isolate public interfaces when packaging an Amplifier workflow. Boundary planning requires separating the core deterministic logic from the thin adapters consumed by pipelines, CLIs, or agent tools.

Shelling out to a public CLI limits direct programmatic integration and introduces process overhead. Importing a shared Python library is preferred when consumers need deterministic constants or deep composability within the same repo.