Reference and project direction¶
The user guides describe released behaviour. The repositories retain detailed contracts, implementation evidence, and architectural decisions for contributors and people building on the engine.
Build your own engine client¶
forge-template is also a Python package with a public API for component
discovery, validation, planning, and in-memory rendering:
uv add "forge-template==0.4.1"
Add it to the project implementing your client. It is not a required dependency of projects generated with the CLI. Start with the public engine API and the runnable downstream client example. Your client owns filesystem writes and command execution; the engine returns validated content in memory.
Organisation defaults and constraints can be resolved by a downstream
client before calling the engine. create-forge itself does not load
organisation policy files, and the engine does not support arbitrary
file-overlay or plugin directories.
Technical reference index¶
| Topic | Source |
|---|---|
| CLI input and prompting | CLI conventions |
| Engine selection and options | Component selection |
| Engine package compatibility | Engine resolution |
| Generation requests | ProjectSpec |
| Catalogue metadata | Component manifests |
| Organisation clients | Policy and constraints |
| Independent clients | Reference client boundary |
| Shared generated tooling | Foundation guarantees |
| Working across repositories | Contributor workflow |
| Why decisions were made | CLI ADRs, template ADRs |
Technical documents on main may describe accepted future contracts. Check
their status sections before treating an interface as a released feature.
What's next¶
The Foundation and Data Science roadmaps are complete. The current release keeps Copier as the default and exposes the engine through an opt-in preview. Making the engine the default is a planned direction, with no scheduled release or supported migration available yet.
Filed roadmap packs now track the Engine-Default Cutover and Streamlit Archetype. These are open plans, not available features. Engine-native updates and continued legacy Copier updates must work before the default switch. Streamlit can proceed after the cutover contracts are accepted; adoption stays on the supported preview path if the cutover has not shipped.
Follow CLI open work, template open work, and the CLI and template releases. The completed Foundation and Data Science roadmaps remain available as historical records.
Suggest a capability or guide to help shape future work.
Missing an example or found an unclear step? Send documentation feedback.