Skip to content

lumioguard CC

Keep AI-written code clean as the project grows.

Coding agents write fast, but the code they leave behind gets complex, duplicated and tangled over time. lumioguard CC sits in the agent loop, checks every change for problems, and blocks the agent from finishing until the code is clean. The result is a codebase that stays maintainable and smaller, so agents read fewer tokens on every future task.

Get started Use it with coding agents

Common tasks

  • Check a change

    Compare your work with the last commit, or your branch with main, and see what the change added.

    Check your changes

  • Use it with coding agents

    Claude Code, Codex and other agents run the check on their own changes and fix what they made worse before they finish.

    Coding agents

  • Check pull requests

    Run the same check in CI. A pull request fails only on the findings it adds.

    Pull requests and CI

  • Reduce existing findings

    List the functions and files with the most findings, fix them in small steps, and compare the numbers before and after.

    Clean up existing code

How it works

flowchart LR
    CODE["Your code"] --> MEASURE["Measure<br/>complexity, size,<br/>duplication, structure"]
    MEASURE --> COMPARE["Compare with<br/>an earlier version"]
    COMPARE --> OLD["Already there<br/>reported, does not fail"]
    COMPARE --> NEW["New or worse<br/>can fail the check"]
  1. Measure. It reads every source file and records a number for each function and file.
  2. Compare. It matches those numbers against a Git commit or a saved snapshot.
  3. Decide. The check passes, fails on findings that are new or worse, or reports that part of the code could not be analyzed. Code it could not analyze never counts as a pass.

The terms used in the results

Where to start

  • New to lumioguard CC

    What it measures, what the results mean, and what a passing check does not tell you. No coding needed.

    What is lumioguard CC?

  • Developers

    Install the binary and run the first check.

    Quick start

  • Looking something up

    Every rule, command, setting and report field.

    Rules · Reference

What it does and does not do

Runs on your machine. It does not upload code, call an AI model or need an account.

Same input, same output. A given version of the tool reports the same numbers on a laptop, in CI and inside an agent.

Incomplete analysis does not pass. If a file cannot be read, the result is incomplete, not passed.

Does not run your code. It reads source files. It does not start your program or your tests.