AI Workflow
===========

FRIDOM is intended to support AI-assisted development, especially for analysis,
planning, refactoring, testing, and documentation work.

Principle
---------

AI systems should follow the same project rules as human contributors. The goal
is not to create a separate workflow, but to make the standard workflow readable
and enforceable for both.

Required lookup order
---------------------

Before changing code, AI-assisted workflows should inspect sources in this
order:

1. the Development documentation
2. the relevant user-facing documentation
3. the existing tests
4. the relevant source code

This reduces the risk of introducing changes that contradict documented or
tested behavior.

Initial agent roles
-------------------

The initial AI workflow should be built around a small set of focused roles:

- planner
- codebase analyst
- refactorer
- reviewer

Optional specialized roles can be added later if they are justified by repeated
use.

Expected behavior
-----------------

AI-assisted contributors should:

- make small, bounded changes
- explain the purpose of changes clearly
- verify relevant tests
- update documentation when needed
- avoid guessing when behavior is unclear

AI systems should stop and ask for guidance when:

- the intended behavior is ambiguous
- public API changes are likely
- architecture tradeoffs are unclear
- existing tests and documentation disagree
- the required rewrite boundary is not well defined

Direction
---------

The AI workflow will be expanded after the development foundation is in place.
This includes defining agent-specific prompts, rule files, and lookup helpers.