[docs] Add AI contribution policy

Document expectations for AI-assisted contributions in a dedicated
policy file. The guidance keeps contributors accountable for
verification, narrow diffs, confidentiality, and licensing.

Testing: not run (documentation-only change)
Fixes: #5790
Assisted-by: OpenAI Codex
This commit is contained in:
Behdad Esfahbod committed 2026-03-12 02:48:26 -03:00
1 parent 07a9ad2ed1
commit d433fc3593
1 file changed
+43
+43
View File
@@ -0,0 +1,43 @@
# Code of AI Conduct
This project accepts AI-assisted contributions, but contributors remain fully
responsible for the changes they submit.
## Requirements
- Use AI as a tool, not as an authority. Verify behavior, correctness, and
compatibility yourself before submitting changes.
- Keep changes narrow. Do not use AI to generate broad rewrites, drive-by
refactors, or style-only churn.
- Preserve HarfBuzz's public API and ABI unless a change explicitly calls for
it.
- Follow the repository guidance in [`AGENTS.md`](AGENTS.md),
[`README.md`](README.md), [`BUILD.md`](BUILD.md), [`TESTING.md`](TESTING.md),
and related local documentation.
- Do not submit generated code, text, or tests that you do not understand.
- Do not fabricate benchmarks, bug reports, test results, or reproduction
steps.
- Do not paste private code, non-public fonts, credentials, tokens, or other
confidential material into external AI systems.
- Respect licensing and attribution requirements. Do not submit AI-generated
content that may have unclear provenance or incompatible licensing.
## Pull requests
- Disclose significant AI assistance in the pull request description.
- Describe what was verified locally, including the exact build or test commands
you ran.
- Call out uncertainty, skipped validation, or areas that need extra review.
- Be prepared to revise or discard AI-generated changes that do not meet the
project's standards.
## Maintainer expectations
- Review AI-assisted contributions by the same technical standards as any other
contribution.
- Prefer reproducible fixes, focused diffs, and adequate tests over volume.
- Reject changes that appear to be unreviewed AI output, including changes that
add unnecessary complexity, invent behavior, or ignore repository guidance.
In short: if AI helped produce a change, a human contributor must still own the
result end to end.