Hi all, Thank you for the vibrant discussion. I align with the points raised above. +1 to start a vote.
Given that the AGENT.md initiative is straightforward and has reached consensus, and considering its potential to immediately enhance the developer experience, I propose we proceed with its implementation without waiting for the formal vote on this FIP. -------------------------------- Hi @Vaibhav Kumar, since you initiated the discussion on AGENT.md, would you be interested in taking ownership of this issue and submitting a pull request? ------------------------------------ Hi Junbo, Regarding the REST API and CLI, here is some design context for your consideration. We recently discussed this in FIP-32, where we plan to introduce a Rust-based Multiprotocol Gateway to support multiple protocols, including REST API. So this deserves a dedicated FIP but may rely on FIP-32. Additionally, we intend to merge the fluss-rust repository back into the main Fluss repository. This consolidation will allow us to integrate the CLI and Gateway directly into the main codebase by leveraging the Rust client. Best regards, Jark On Thu, 19 Mar 2026 at 19:13, Yang Wang <[email protected]> wrote: > > Hi Mehul, > > Thank you for the thorough feedback and the +1 on Phase 1! These are > extremely well-researched suggestions — all zero-cost and immediately > actionable. Let me respond to each one. > > 1. CodeRabbit for AI-Assisted PR Review** > > This is a great fit for Dimension 1's Phase 3 tooling. Free Pro tier for > public repos, zero ASF runner consumption, context-aware line-by-line > review with configurable rules via `.coderabbit.yaml` — the value > proposition is compelling. We could configure Fluss-specific rules for > architecture boundaries, ASF license headers, and coding conventions, which > directly complements the AI Contribution Guidelines in Phase 2. > > I'll add this as a concrete candidate in Phase 3's development ecosystem > tooling. > > 2. kapa.ai Doc Bot** > > I think this aligns well with Dimension 2's broader goal of making Fluss > more accessible to AI-powered workflows. While the FIP's Dimension 2 > roadmap focuses on REST API → CLI → Skills for programmatic operation, an > AI-powered documentation experience is a valuable complementary direction > that lowers the barrier for new users. The free open source program (up to > 10k questions/month) and the fact that projects like Polars, LangChain, and > Nuxt are already using it makes this worth exploring. > > 3. GitHub AI Issue Labeler + Moderator** > > This fits naturally into Dimension 1's Phase 3 tooling. Being GitHub-native > with zero extra API keys needed makes it the lowest-friction option among > all the suggestions. Auto-labeling issues by component (fluss-lake-iceberg, > fluss-flink, fluss-server, etc.) would be immediately useful for triage, > and the AI moderator can help maintain community health as the project > grows. I'll add this to the Phase 3 tooling exploration. > > 4. OpenSSF Scorecard + Dependabot** > > Agreed — these are proven security best practices that many ASF projects > already use, and adopting them at zero cost is a no-brainer. While not > AI-specific, they strengthen the overall development infrastructure that > supports AI-assisted workflows. I'll track this as an actionable > improvement alongside the FIP rollout. > > 5. Module-Level AGENTS.md Priority** > > Your proposed starting point — fluss-server, fluss-lake-iceberg, and > fluss-flink-common — is spot on. These are the three most architecturally > complex modules where AI agents are most likely to make mistakes without > proper context. I'll update section 3.1 of the FIP to include this > prioritization guidance. > > One open question for the community > > Several of these suggestions (CodeRabbit, kapa.ai) involve integrating > third-party commercial services — free for open source, but commercial > nonetheless. I'm not sure whether ASF has specific policies or precedents > around installing third-party GitHub Apps or embedding third-party SaaS > widgets on project websites. If anyone in the community has experience or > knowledge about ASF's stance on this, I'd appreciate the input so we can > move forward with clarity. > > Thanks again for the detailed and actionable suggestions, Mehul. This is > exactly the kind of concrete input that makes the FIP better. > > Best regards, > Yang Wang > > > Mehul Batra <[email protected]> 于2026年3月19日周四 17:10写道： > > > Hi Yang, > > > > Thank you for driving this FIP. +1 on landing Phase 1 as proposed. > > > > A few additional suggestions, all zero-cost for the project: > > > > 1. CodeRabbit for AI-Assisted PR Review > > > > CodeRabbit [1] offers its full Pro tier free forever for public > > repositories. It provides context-aware, line-by-line code review on every > > PR, PR summaries, incremental reviews on new commits, and interactive chat > > via @coderabbitai in PR comments. It runs on CodeRabbit's own > > infrastructure, so it doesn't consume ASF GitHub Actions runners. > > > > Setup is just installing the CodeRabbit GitHub App and adding a > > .coderabbit.yaml to configure Fluss-specific review rules (architecture > > boundaries, ASF license headers, naming conventions, test expectations). > > > > 2. "Ask AI" Doc Bot for the Fluss Website > > > > On Dimension 2 (AI-Friendly Product), I'd suggest exploring kapa.ai [2] > > for > > an AI-powered "Ask AI" widget on the Fluss documentation site. Kapa offers > > a free open source program (up to 10k questions/month) for qualifying > > projects. It ingests your docs and lets users ask natural language > > questions directly on the website, similar to what Polars, LangChain, and > > Nuxt already use. > > > > Fluss qualifies (Apache-licensed, non-commercial, publicly available). This > > would lower the barrier for new users trying to understand Fluss concepts, > > configuration, and the lakehouse integration without digging through pages > > manually. > > > > 3. GitHub AI Issue Labeler + Moderator for Triage > > > > GitHub recently released two free AI-powered Actions using the GitHub > > Models inference API [3]: an AI assessment comment labeler that > > auto-categorizes issues (bug, feature request, question, etc.) and an AI > > moderator that detects spam and AI-generated low-quality content. Both use > > the workflow's GITHUB_TOKEN with models:read permission, so no extra API > > key is needed. > > > > For Fluss this could auto-label issues by component (fluss-lake-iceberg, > > fluss-flink, fluss-server, etc.) and type, reducing manual triage overhead > > as the project grows. > > > > 4. OpenSSF Scorecard + Dependabot for Security > > > > OpenSSF Scorecard [4] is a free GitHub Action that runs automated security > > health checks on every push or on a schedule. It checks branch protection, > > dependency update tooling, SAST presence, signed releases, code review > > practices, vulnerability disclosure, and more, then surfaces findings in > > the GitHub Security tab. > > > > Combined with Dependabot (already free for public repos), this gives Fluss > > automated dependency vulnerability alerts and security posture scoring at > > zero cost. Many ASF projects already use both. > > > > 5. Module-Level AGENTS.md Content > > > > For the module-level AGENTS.md, I'd suggest starting with the three most > > architecturally complex modules: fluss-lake-iceberg (tiering writer > > architecture, shading constraints, Iceberg catalog rules), > > fluss-flink-common (Flink connector contracts, multi-version > > compatibility), and fluss-server (coordinator logic, replication, ZooKeeper > > interactions). > > > > Best regards, > > Mehul > > > > [1] https://www.coderabbit.ai/open-source > > [2] https://docs.kapa.ai/kapa-for-open-source > > [3] > > > > https://github.blog/changelog/2025-09-05-github-actions-ai-labeler-and-moderator-with-the-github-models-inference-api/ > > [4] https://github.com/ossf/scorecard-action > >
