Classifier.dev launches semantic-ci-lint for PR reviews

// 1h agoOPENSOURCE RELEASE

Classifier.dev launches semantic-ci-lint for PR reviews

semantic-ci-lint is an open-source skill and CI workflow from classifier.dev that checks changed git diff hunks against codebase conventions written in prose, such as error-handling guidelines and PII logging restrictions. Instead of passing whole source files into generative LLMs, the tool uses a lightweight Python script with git diff -U0 to evaluate added lines against single-convention prompts via classifier.dev's keyless zero-shot classification API.

// ANALYSIS

Enforcing codebase prose conventions through zero-shot classification on diff hunks is significantly cheaper, faster, and more reliable than deploying chatty generative LLMs to review PRs.

–Targeted diff inspection: Classifying isolated diff hunks alongside their file paths captures sufficient context for semantic analysis without dragging full files or unrelated context into the evaluation.
–Calibrated confidence thresholds: Enforcing an advisory model where comments only trigger at 0.9+ confidence (yielding 82–92% empirical accuracy) shields engineering teams from hallucinated nitpicks and bot muting.
–Natural complement to AST linters: While compilers and tools like ESLint or Ruff handle deterministic syntax checks, semantic classification cleanly bridges the gap for human-written style guides and architectural rules that defy regex.
–Frictionless CI deployment: Leveraging classifier.dev's keyless free tier (up to 3,000 classifications per minute) allows teams to spin up semantic linting directly in GitHub Actions with zero API credential management.

// TAGS

ci-cdgithub-actionscode-reviewsemantic-cilintingagent-skillsdevopszero-shot-classification

DISCOVERED

1h ago

2026-09-19

PUBLISHED

1h ago

2026-09-19

RELEVANCE

7/ 10

AUTHOR

michael_chomsky

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

RESEARCH1h ago

ScientistTwo automates end-to-end scientific research

Developed by Google Cloud AI Research and the University of Waterloo, ScientistTwo is an autonomous multi-agent system designed to handle the complete scientific research lifecycle at scale. Spanning ten AI domains, the framework autonomously generates hypotheses, executes code in sandboxes, runs automated ablation studies, drafts academic manuscripts, and performs simulated peer-review cycles where rebuttal agents run new empirical experiments to address critiques.

OPEN SOURCE2h ago

MiniMax open-sources MiniMax Code coding agent

MiniMax has announced the open-source release of MiniMax Code, bringing its agentic software engineering capabilities into developers' local environments via a terminal CLI and desktop workspace. By opening up the project, MiniMax aims to give developers full visibility into the tradeoffs of autonomous coding agents, enabling teams to execute commands locally, coordinate multi-agent tasks, and customize agent behaviors to fit their own development toolchains.

LAUNCH2h ago

Classifier.dev launches free zero-shot classifier beating Jev

Classifier.dev offers a free, highly accessible tool for zero-shot text classification that can be used directly within environments like Grok, Claude Code, Codex, and ChatGPT. Positioned as an alternative to the Jev classification model, it utilizes small large language models (LLMs) in a "smart mode," which the creators claim delivers superior performance compared to the original model.