Newshot

Show HN: Lintsure – Free local code check with rules from OWASP, EU AI Act

Original text(via HN AI)

Lintsure

Catch your AI coding agent's mistakes before they ship. Open-source guardrails for Claude Code, Cursor, Codex, VS Code and Gemini CLI. Lintsure keeps your agent working to the rules you set, and checks the code it writes for security and privacy problems against the standards you connect: OWASP, GDPR, the EU AI Act and more. Runs on your machine · Same input, same result · CLI, MCP server and GitHub Action

npm version
npm version
Latest release
Latest release
Node version
Node version
Important: Lintsure is a readiness tool, not an audit, a certification or legal advice. Its rules reflect our reading of the publicly available text of each framework, and they check code only. Work outside code, such as policies, training, contracts and evidence, still has to be done. Have a qualified person review the results before any audit, certification or decision. See DISCLAIMER.md.

What it does

Lintsure sits between your AI coding agent and your codebase. It does three things:

  • Checks the code your agent writes. 74 rules look for problems that AI-written code often has: secrets committed to source, databases without row level security, personal data written to logs, AI output used without a human check, agent configs that run remote code, and more. It reads JavaScript, TypeScript and Python, plus config and secret checks for every file.
  • Maps each finding to the standards you connect. Six frameworks are built in: OWASP Top 10, OWASP LLM Top 10, CWE Top 25, GDPR, the EU AI Act and the EU Cyber Resilience Act. You pick which ones apply in lintsure.config.yaml.
  • Keeps your agent on your rules. The MCP server lets the agent ask which checks apply before it writes code (plan_checks) and scan its own work before it says it's done. The Claude Code plugin adds Drift Guard, which holds the agent to the rules you write in .lintsure/rules.yml.

Every finding shows:

  • the file and line
  • a status: Gap, Partial, Met or Outside code
  • every control it touches, across every framework you check
  • a suggested fix and a plain auditor note

It also lists what code can't prove (access reviews, training records, contracts), and it can write a multi-page assessment report with a human review and sign-off page (--report).

Common questions it answers: Is there a security scanner for code written by Claude Code or Cursor? How do I stop my AI agent from committing secrets? Is there an MCP server that checks code against OWASP or GDPR? How do I make my coding agent follow project rules?


        ✓ 3 findings · 2 gap · 1 partial · 13 rules met · 22 outside code
partial  src/billing.ts:4  Data sent over plain HTTP
gap      src/config.ts:2   Secret or private key committed to source
gap      src/users.ts:5    Personal data written to logs
      

Built to inform, not to decide

  • Never uploads your code. The engine has no network code. A test fails the build if any appears.
  • Same input, same result. The Checker is a fixed rule engine, not a language model.
  • No fake green ticks. Only the Checker can mark a control Met, and Met only means the rules found no problem in the files they read.
  • Never decides for you. Every report and every MCP reply ends with the disclaimer and the next steps a person still needs to take. The MCP server tells coding agents not to call you compliant, certified or ready, not to make decisions, and to ask before changing anything. Tests check that Lintsure's own wording never claims a verdict.

What it doesn't do

We'd rather say this up front than have you find out later:

  • Not a pentest. It reads source files. It doesn't run your app, send requests to a live site or test a deployed server.
  • Not a full code analyzer. Rules are pattern checks on lines and statements, plus a few checks across files. They don't follow data through your whole program, so some findings are marked Partial for a person to confirm.
  • Not a dependency scanner. It flags risky install patterns and unpinned versions, not known vulnerabilities in the packages you use.
  • Not an audit or legal advice. A Met control means the rules found no problem in the files they read, nothing more.

Using it with an AI agent: Lintsure is a defensive review of code you own. If you ask an agent to run it, say that ("review my own repo against OWASP Top 10 and fix what you find"), which describes the work accurately.

Try it

Status: Beta. All rules are in Beta: their mappings are our own reading of the public texts, not independently reviewed.

        npx lintsure scan
      

Add it to the coding agents this project uses (Claude Code, Cursor, VS Code, Gemini CLI). It shows the change first and backs up any file it edits:


        npx lintsure setup          # preview
npx lintsure setup --yes    # write it
npx lintsure status         # check Node, config, rules and agents
      

Or add it yourself in Claude Code: claude mcp add lintsure -- npx -y @lintsure/mcp.

From source: git clone, npm install, npm test, then node packages/cli/dist/src/index.js scan /path/to/your-app.

Then ask: check this repo for compliance. Setup for Cursor, Codex and Devin Desktop (formerly Windsurf), every CLI flag, the report format and the full rules reference are in the docs.

Run it on every pull request with lintsure/action.

Docs offline: the @lintsure/core package includes plain Markdown docs (docs/rules.md, docs/frameworks.md, docs/agents.md), generated from the rules themselves, so a coding agent can read them from node_modules/@lintsure/core/docs without going online.

What's inside

PathWhat
packages/coreFinding schema, 74 rules, framework knowledge, scan, config reader, report writer, policy fingerprint, pack loader, safety wording
packages/cliThe lintsure command: scan , baseline , init , info , setup , status , export
packages/mcpMCP server: scan , plan_checks , classify , explain_framework , explain_rule
fixtures/Planted problems, near misses and a config example used by the tests
scripts/mcp-smoke.mjsStarts the MCP server: every reply carries the disclaimer, bad input is rejected, the tool list stays small
scripts/cli-smoke.mjssetup and status in a temporary project
scripts/check-sources.mjsChecks the real-incident links on rules still answer (run before a release)

Get it from the official places only

Lintsure is published only as lintsure, @lintsure/core and @lintsure/mcp on npm, from github.com/lintsure, and at lintsure.com. A security tool is a good target for lookalikes: a repo or package with a similar name is not ours.

Help Lintsure grow

Lintsure is free and built in the open. If it helped you:

  • ⭐ Star this repo, so more teams can find it
  • Share it with a founder or engineer who's facing an audit
  • Send a review if you run it at your company. Good or bad, we read every one
  • Request a demo of the team platform (history, PR checks, evidence log, expert review)

Get help

  • Security issues: see SECURITY.md. Please don't open a public issue.
  • Bugs and ideas: open an issue

What's new

See CHANGELOG.md or lintsure.com/changelog.

Contributing

See CONTRIBUTING.md. Every rule needs a planted example, a near miss, and a link from the framework knowledge; npm test checks all of it.

Everyone taking part follows the code of conduct. Our accessibility commitment, and how to report a barrier, is in ACCESSIBILITY.md.

License

Apache-2.0. Contributions are accepted under the DCO: sign off your commits with git commit -s.

The text above was extracted automatically from the source page and may miss images, tables or code blocks. The original is authoritative.

Published by HN AI
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗