Apache Fory is a performance-critical foundational serialization framework with cross-language compatibility requirements. AI tools are welcome as assistants, but project quality, legal safety, and maintainability standards are unchanged.
The key words MUST, MUST NOT, REQUIRED, SHOULD, and MAY are interpreted as described in RFC 2119.
- AI tools MAY assist contribution work.
- AI tools MUST NOT replace contributor accountability.
- The human submitter is responsible for correctness, safety, performance, and maintainability of all submitted changes.
- License/provenance confirmation: contributors MUST confirm submitted material is legally compatible and traceable, and MUST comply with ASF Generative Tooling Guidance.
- AI-assisted code MUST be reviewed carefully by the contributor line by line before submission.
- For substantial AI assistance, contributors MUST complete the required AI review loop in Section 5 before opening or updating the PR for maintainer review.
- Contributors MUST be able to explain and defend design and implementation details during review.
- AI-generated code can appear plausible while still containing correctness, protocol, performance, maintainability, or licensing problems.
- When AI materially contributes technical content, the contributor's main responsibility shifts toward rigorous review, correction, and validation, not just drafting.
- If contributors skip that review work, the effective burden is transferred to maintainers, which is not an acceptable review model for this project.
- The required AI review loop exists to make contributors complete that review and correction work before requesting maintainer time.
For substantial AI assistance, PR descriptions MUST include a short AI Usage Disclosure section.
For minor or narrow AI assistance, full disclosure is not required.
The PR template keeps this section intentionally short and links to Section 9 for the complete checklist.
Definition of substantial AI assistance:
- Substantial means AI materially influenced technical content, not only writing style.
- Contributors MUST mark AI assistance as substantial (
yes) if any of the following apply:- AI generated or rewrote non-trivial code/test logic (even for a small change or a single function).
- AI-generated or AI-refactored content is about 20 or more added/changed lines in aggregate.
- AI materially influenced API, protocol, type mapping, performance, memory, or architecture decisions.
- AI produced substantive technical text used in PR rationale (beyond grammar/translation cleanup).
- Contributors MAY mark substantial AI assistance as
nofor minor or narrow assistance only, such as spelling/grammar fixes, formatting, trivial comment wording edits, or other non-technical edits with no behavior impact.
Required disclosure fields:
- Whether substantial AI assistance was used (
yesorno) - Scope of assistance (for example: design drafting, code drafting, refactor suggestions, tests, docs)
- Affected files or subsystems (high-level)
- AI review summary (self-review completed, AI review loop status, and final result)
- Final AI review artifacts (embedded screenshots or links showing the final clean AI review results from both fresh reviewers on the current PR diff or current HEAD after the latest code changes)
- Human verification performed (checks run locally or in CI, and results reviewed by the contributor)
- Provenance and license confirmation (see Section 6)
Standard disclosure template (recommended):
AI Usage Disclosure
- substantial_ai_assistance: yes
- scope: <design drafting | code drafting | refactor suggestions | tests | docs | other>
- affected_files_or_subsystems: <high-level paths/modules>
- ai_review: <line-by-line self-review completed; summarize the two-reviewer loop and final no-further-comments result>
- ai_review_artifacts: <embedded screenshots or links showing the final clean review results from both fresh reviewers on the current PR diff or current HEAD after the latest code changes>
- human_verification: <checks run locally or in CI + pass/fail summary + contributor reviewed results>
- performance_verification: <N/A or benchmark/regression evidence summary>
- provenance_license_confirmation: <Apache-2.0-compatible provenance confirmed; no incompatible third-party code introduced>
To protect privacy and enterprise security:
- Contributors are NOT required to disclose model names, provider names, private prompts, or internal workflow details.
- Maintainers MAY request additional clarification only when necessary for legal or technical review.
The following MUST be human-authored (translation and grammar correction tools are acceptable):
- Review responses
- Design rationale and tradeoff discussion
- Risk analysis and production impact explanation
Generated filler text, evasive responses, or content that does not reflect contributor understanding may result in PR closure.
For non-trivial changes (especially architecture/protocol/performance-sensitive work), contributors SHOULD align scope in an issue or discussion before implementation.
This expectation applies to all contributors, whether AI-assisted or not. For AI-assisted non-trivial work without prior alignment, maintainers MAY request scope alignment before continuing review.
Every AI-assisted PR MUST provide verifiable evidence of local or CI validation:
For substantial AI assistance, every PR MUST also provide verifiable evidence of a completed AI review loop before maintainer review:
- The contributor personally performs a line-by-line self-review first and fixes all issues found before requesting AI review.
- The contributor then runs two fresh AI review agents on the current PR diff or current HEAD after the latest code changes:
- one reviewer MUST use
.claude/skills/fory-code-review/SKILL.md - one reviewer MUST NOT use that skill
- one reviewer MUST use
- The contributor addresses all actionable comments from both reviewers, reruns both reviewers on the updated diff, and repeats this loop until both reviewers report no further actionable comments.
- The PR body MUST include the final clean AI review result from both reviewers plus screenshot evidence in the
AI Usage Disclosure. - If the contributor cannot produce this evidence, the PR is not ready for maintainer review.
Definitions for AI review evidence:
- Fresh AI review agent means a new clean-context review session started on the current diff after the latest code changes. Reusing an old reviewer thread as the final review evidence is not sufficient.
- Final clean AI review result means the last rerun of both reviewers on the current PR diff or current HEAD, with no unresolved actionable comments remaining.
- Screenshot evidence must show, for each reviewer, the reviewer identity or workflow label, the reviewed diff/commit or PR state, and the clean no-further-actionable-comments result. Persisted links with equivalent information MAY be used when screenshots are impractical.
Definition of adequate human verification:
-
The contributor personally runs the relevant checks locally or in project CI and reviews the results.
-
Verification includes concrete evidence (exact commands and pass/fail outcomes), not only claims.
-
Verification covers the changed behavior with targeted tests where applicable.
-
For protocol or performance-sensitive changes, verification includes the required compatibility tests and/or benchmark/regression evidence.
-
Confirmation that contributor performed line-by-line self-review of AI-assisted code changes
-
Build/lint/test checks run locally or in CI
-
Targeted tests for changed behavior
-
Results summary (pass/fail and relevant environment context)
Additional REQUIRED checks for Fory-critical paths:
- Protocol, type mapping, or wire-format changes:
- Update relevant docs under
docs/specification/** - Add or update cross-language compatibility tests where applicable
- Update relevant docs under
- Performance-sensitive changes (serialization/deserialization hot paths, memory allocation behavior, codegen, buffer logic):
- Provide benchmark or regression evidence
- Justify any measurable performance or allocation impact
Claims without evidence may be treated as incomplete.
Contributors MUST follow ASF legal guidance and project licensing policy, including:
- ASF Generative Tooling Guidance
- ASF third-party licensing requirements
- Apache-2.0 compatibility obligations
Contributors MUST ensure:
- No incompatible third-party code is introduced
- Reused material has compatible licensing and required attribution
- AI-generated output does not include unauthorized copyrighted fragments
If provenance is uncertain, contributors MUST remove or replace the material before submission. Maintainers MAY request provenance clarification when needed.
Maintainers MAY close or return PRs that materially fail project standards, including:
- Contributor cannot explain key implementation logic
- Missing required disclosure for substantial AI assistance
- Missing required AI review loop evidence, final clean reviewer outputs, or screenshot artifacts in the PR body
- Missing or inadequate human verification evidence for changed behavior
- Redundant implementation of existing utilities without clear necessity
- Introduction of dead code, unused helpers, or placeholder abstractions without justification
- Protocol or performance claims without reproducible evidence
- Large unfocused changes with unclear scope or ownership
This is not a ban on AI usage; it is a quality and maintainability gate.
Before merge, maintainers MAY request:
- Additional tests, benchmarks, or spec updates
- PR split into smaller verifiable commits
- Clarification of technical rationale, provenance, or licensing
- Rework of sections that do not meet standards
Maintainers MAY close PRs that remain non-compliant after feedback. For substantial AI-assisted PRs that omit the required final AI review evidence, maintainers MAY close the PR directly without performing review on the contributor's behalf.
Any long-term contribution restrictions MUST follow Apache project governance and community process, and SHOULD be documented with clear rationale.
This is the canonical checklist for the PR template AI section.
- Substantial AI assistance was used in this PR:
yes/no - If
yes, I included the standardizedAI Usage Disclosureblock below. - If
yes, I can explain and defend all important changes without AI help. - If
yes, I reviewed AI-assisted code changes line by line before submission. - If
yes, I completed line-by-line self-review first and fixed issues before requesting AI review. - If
yes, I ran two fresh AI review agents on the current PR diff or current HEAD after the latest code changes: one using.claude/skills/fory-code-review/SKILL.mdand one without that skill. - If
yes, I addressed all AI review comments and repeated the review loop until both ai reviewers reported no further actionable comments. - If
yes, I attached screenshot evidence of the final clean AI review results from both fresh reviewers on the current PR diff or current HEAD after the latest code changes in this PR body. - If
yes, I ran adequate human verification and recorded evidence (checks run locally or in CI, pass/fail summary, and confirmation I reviewed results). - If
yes, I added/updated tests and specs where required. - If
yes, I validated protocol/performance impacts with evidence when applicable. - If
yes, I verified licensing and provenance compliance.
AI Usage Disclosure (only when substantial AI assistance = yes):
AI Usage Disclosure
- substantial_ai_assistance: yes
- scope: <design drafting | code drafting | refactor suggestions | tests | docs | other>
- affected_files_or_subsystems: <high-level paths/modules>
- ai_review: <line-by-line self-review completed; summarize the two-reviewer loop and final no-further-comments result>
- ai_review_artifacts: <embedded screenshots or links showing the final clean review results from both fresh reviewers on the current PR diff or current HEAD after the latest code changes>
- human_verification: <checks run locally or in CI + pass/fail summary + contributor reviewed results>
- performance_verification: <N/A or benchmark/regression evidence summary>
- provenance_license_confirmation: <Apache-2.0-compatible provenance confirmed; no incompatible third-party code introduced>
This policy complements, but does not replace, existing ASF and Apache Fory governance, contribution, and legal policies. If conflicts arise, ASF legal and project governance rules take precedence.