Code review is the new bottleneck for development teams drowning in pull requests, and a new open-source tool called fks2g wants to help you triage smarter. Dropped on Hacker News by developer kmdupr33, this CLI utility uses LLM-backed analysis to score how closely each file change needs to be reviewed—essentially building a risk matrix before you even crack open the diff.

How FKS2G Works

The tool collects five distinct data points to inform its review recommendations. First, it calculates cosine similarity between file-name embeddings and configurable project text sources like documentation or specs. Second, an LLM judges whether files likely to change soon based on those same source documents—essentially asking "does this code match where the team is headed?" Third, recent bug-fix commits get classified by another LLM pass to surface patterns in past instability. Fourth, git history provides raw file change frequency metrics. Finally, all this evidence feeds into a concluding LLM risk assessment that spits out high/medium/low ratings with reasoning. The CLI accepts OpenAI API keys and can target specific GitHub repositories for analysis. In the example shown, it analyzes files from the most recent React commit using GPT-5.5, feeding in file paths via standard git output piping. The tool theoretically supports Google, Anthropic, and AWS Bedrock models as well, though kmdupr33 notes they've only personally tested with OpenAI implementations.

Real Output: Analyzing the React Repo

The example output reveals how fks2g thinks about risk. For package.json—a file changed 763 times historically—the tool flagged it as "high" risk because frequent changes increase collision probability with ongoing work, and there was no source-document signal suggesting imminent updates. Meanwhile, yarn.lock came in at "medium" despite 415 historical changes, with the LLM noting recent bug-fix activity (a FlightReply fix for FormData handling) created some collision risk but no clear upcoming-work indicators from project docs. More interesting is how the tool handles rarely-changed files. The ReactFlightDOMServerNode.js file in react-server-dom-webpack had only 12 historical changes, yet still rated "medium" because recent bug-fix activity and broad RSC (React Server Components) relevance made modifications moderately risky. Critically, fks2g pulls actual GitHub issue references when available—it cited issues #36491, #36430, and others as source signals for why that particular file warranted attention despite its historical stability.

Caveats: This Is 0.X Software

Let's be clear about what you're getting here. The project explicitly warns it's "as ready for prime-time usage as its name suggests," which is a fun way of saying early-alpha experimental tooling. kmdupr33 admits they've only kicked the tires with OpenAI models, and while Google/Anthropic/Bedrock support is theoretically baked in, there's no telling how well those integrations actually perform without community testing.

Key Takeaways

  • FKS2G scores files high/medium/low for review intensity using five LLM-backed signals including embeddings, change prediction, bug-fix patterns, and git history.
  • The tool can cite actual GitHub issues as context for why certain files warrant attention despite low historical change frequency.
  • Version 0.X status means production deployment requires significant vetting—currently best suited for experimentation and evaluation.

The Bottom Line

The core idea is solid—code review triage based on historical risk signals plus forward-looking LLM judgment could genuinely reduce cognitive load for overstretched teams. But at version 0.X with minimal testing beyond one developer's laptop, this needs serious vetting before production use. Worth watching, probably not worth deploying yet.