Skip to content

AI_POLICY.md

Latest commit

 

History

History
75 lines (62 loc) · 4.05 KB

AI_POLICY.md

File metadata and controls

75 lines (62 loc) · 4.05 KB
Raw

AI Usage & Contribution Policy

The Zenith Example project welcomes the responsible use of AI coding assistants (such as Claude Code, Antigravity, Cursor, GitHub Copilot, and LLMs) to augment human development. To preserve codebase quality, architectural clarity, and maintainer time, all contributors must adhere to the following principles:

1. Human-in-the-Loop & Full Accountability

  • Human Ownership: All AI-assisted code, documentation, and issues must be thoroughly understood, reviewed, and edited by a human contributor before submission.
  • You Are Responsible: By submitting a pull request, you accept full personal accountability for the submitted changes, including bug fixes, type safety, test coverage, and long-term maintainability.

2. Verification Gate Before Submitting

  • Mandatory Local Testing: Never submit AI-generated code without running the full test and verification suite locally:

    uv run poe test
uv run poe code-quality
uv run prek run --all-files

  • No Speculative Solutions: Do not submit code for environments, platforms, or hardware you cannot test locally. Relying on remote CI pipelines to discover if AI-generated code works is prohibited.

3. No Unattended Bots or PR Floods

  • Automated Agents Restricted: Autonomous, unattended agent scripts are strictly forbidden from directly opening pull requests, submitting issues, or posting bot comments on this repository.
  • Targeted Contributions: Pull requests created with AI assistance should address existing, accepted issues or open discussions. Drive-by PRs addressing non-issues may be closed without review.
  • Search Before Opening (No Duplicates): Before opening a pull request, search open and closed pull requests, issues, and discussions. Do not open a pull request when the same change has already been proposed or is currently in progress. Submitting duplicate pull requests wastes maintainer bandwidth.
  • Contribute at Human Speed, Not AI Speed: AI tools make it trivial to generate large volumes of code and prose quickly, but open source review bandwidth moves at human speed. Do not post rapid-fire pull requests, mass-claim open issues, or flood issue trackers with automated solutions.
  • Do Not Communicate via LLM-Generated Text: Do not rely on AI assistants to speak for you in issues, pull request discussions, or reviews. Maintainers expect direct, thoughtful communication written by the contributor, not raw LLM summaries.

4. Transparent Disclosure

  • PR Disclosure: Contributors using AI coding tools must state the tools used (e.g. Claude Code, Antigravity, Cursor) and the general scope of assistance in the pull request description (see .github/PULL_REQUEST_TEMPLATE.md).
  • Commit Attribution: When AI coding assistants contribute directly to code generation or commits, credit them in Git commit messages using standard Co-authored-by trailers (e.g. Co-authored-by: Claude <noreply@anthropic.com>).
  • Protecting Newcomer Issues: Issues tagged with good first issue are reserved for human contributors learning the codebase. Using AI to generate automated solutions for these issues defeats their educational purpose and is discouraged.

5. Concise Code, Clear Comments, No Slop

  • Default to Zero Inline Comments Unless Explaining 'Why': Comments should default to none. Only add inline comments when explaining a non-obvious rationale, hidden constraint, workaround, or subtle architectural invariant.
  • Remove Obsolete, Verbose, & Diff-Narrative Commentary: Any comments added must be concise and relevant for the current state of the code. Remove any AI-generated comments that describe changes relative to the previous state of the code, narrate what the agent did or did not do, or restate obvious language semantics before opening a PR.
  • Maintain Architectural Invariants: Ensure AI-generated code respects project conventions (PEP 695 generics, slotted dataclasses, pure functions, absolute imports, and layer boundaries defined in AGENTS.md).