vaquarkhan opened a new issue, #17670:
URL: https://github.com/apache/iceberg/issues/17670

   ### Feature Request / Improvement
   
   ### Background
   
   The `[DISCUSS] AI Review comments` thread on [email protected] raised 
that AI-generated review comments are often verbose, sometimes posted without 
human review, and can conflict across different AI tools. While investigating, 
I also found related issues in recent commit metadata on `main`.
   
   ### Findings (last ~2 months on `main`)
   
   -  PRs merged; 9.6% PR show LLM authorship/assist signals; 80 are Dependabot.
   - AI tools are being recorded as `Co-authored-by:` with bot `noreply@` 
addresses.
   - The same tool appears under multiple identities, which fragments 
contributor
     stats and misrepresents provenance. Examples observed:
     - `Claude Opus 4.8 (1M context) <[email protected]>`
     - `Claude Opus 4.8 <[email protected]>`
     - `Claude Opus 5 (1M context) <[email protected]>`
     - `Claude Opus 4.6 (1M context) <[email protected]>`
     - `Claude <[email protected]>`
     - `Codex <[email protected]>`
     - `Cursor <[email protected]>`
   
   `AGENTS.md` currently covers PR-authoring disclosure and the `Generated-by:` 
token well, but has only a single soft line on review comments and nothing on 
co-author identity or AI review-agent safety.
   
   ### Proposal
   
   Expand the AI section of `AGENTS.md` to add:
   
   1. **AI-Assisted Review Comments** -  a human must read, understand, and 
endorse  each AI comment and post it under their own account; keep comments 
concise
      (what/where/fix); at most one AI reviewer per PR; no auto-posting.
   2. **Crediting AI Tools** - disclose AI via the `Generated-by:` token, not 
`Co-authored-by:` bot identities; use one canonical tool name; reserve
      `Co-authored-by:` for humans.
   3. **AI Review Agent Safety** - treat agent-instruction files (`AGENTS.md`, 
CI  agent configs) as security-sensitive; run AI review agents with least 
privilege and no auto-acting on untrusted PR content. (Motivated by documented 
prompt injection via agent-instruction files.)
   4. **Boundaries** - AI-authored PRs must pass the same review, test, and  
security/static-analysis gates as human-authored PRs.
   
   ### Scope
   
   Docs-only change to `AGENTS.md`. A PR implementing this will be linked.
   
   ### Links
   
   - `[DISCUSS] AI Review comments` thread: 
https://lists.apache.org/thread/y93010hlq2m59jjw9lsjbrs38xw8nkdb
   
   
   ### Query engine
   
   None
   
   ### Willingness to contribute
   
   - [ ] I can contribute this improvement/feature independently
   - [ ] I would be willing to contribute this improvement/feature with 
guidance from the Iceberg community
   - [ ] I cannot contribute this improvement/feature at this time


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to