Semlint - Semantic Linting for Modern Codebases

Why Semlint?

🤖

Agentic Backend

Leverages LLM agents via standard CLI tools (like cursor-cli). Semlint handles the orchestration, diffing, and strict JSON output enforcement.

⚡

Cost-Effective Batching

Run all applicable semantic rules in a single LLM pass with --batch mode, dramatically reducing inference costs and wait times.

🌳

Git-Aware Execution

Automatically scopes analysis to your local branch changes (tracked, staged, unstaged, and untracked) or specific git refs.

How It Works

Configure your environment

Create a semlint.json at your project root. Connect it to your preferred agentic backend.

semlint.json

{
  "backend": "cursor-cli",
  "model": "auto",
  "execution": {
    "batch": true
  },
  "backends": {
    "cursor-cli": {
      "executable": "cursor"
    }
  }
}

Define semantic rules

Add JSON rules in your rules/ directory. Use natural language to describe what the agent should look for.

rules/SEMLINT_NAMING_001.json

{
  "id": "SEMLINT_NAMING_001",
  "title": "Ambient naming convention consistency",
  "severity_default": "warn",
  "include_globs": ["src/**/*.ts"],
  "exclude_globs": ["**/*.test.ts", "**/*.spec.ts"],
  "prompt": "Verify naming is consistent with ambient conventions, infer the convention from the surrounding code."
}

Run checks locally or in CI

Semlint filters rules by diff context, builds precise prompts, handles LLM flakiness, and emits standard diagnostic output.

bash

$ semlint check --batch

src/main.ts
  64:1  warn  SEMLINT_NAMING_001  Variable 'rules_dir' uses snake_case; ambient convention is camelCase (e.g. changedFiles, runnableRules). Use 'rulesDir' instead.

✖ 1 problem (0 errors, 1 warning)