AI-First Language Experience
0x0 treats AI help as a normal way to read, explain, and change the codebase.
AI-first does not mean AI can invent behavior. It means every answer, patch,
example, and documentation step must be tied to implemented behavior, checked
documentation, or a clearly named unsupported boundary.
Prompting Boundary
When asking an agent to work on 0x0, include:
- the exact goal;
- the files or roadmap milestone involved;
- whether heavy gates may run;
- whether docs, ADRs, and RFCs must be updated;
- the expected commit boundary.
Agents should read the local source and docs before changing behavior. They
must not copy unsupported external behavior into current 0x0 docs unless
the 0x0 repository has implementation evidence for that behavior.
Implemented Behavior Rule
Use these evidence classes, in this order:
1. source implementation;
2. executable test or gate;
3. release artifact or generated report;
4. reference documentation;
5. public learning documentation.
If a feature has no evidence, document it as unsupported or omit it from the
current behavior path.
Beginner Explanation Rule
Learning-oriented pages should explain the idea first, then show the smallest
current example. Code-facing pages should stay exact and professional.
Safe Command Rule
Default AI workflows should use bounded gates. Heavy gates such as
make selfhost-guard, make release-verify, and ecosystem release checks are
explicit operations, not background validation.