← Home Products · Agent Check Updated

Agent Check: your AI agent checks the code it just wrote

Connect VibeDoctor to Claude Code, Cursor or Codex once. Your agent sends the files it changed to vibedoctor_check, gets each issue back with its line and a fix, and fixes it before you commit. Nothing to download or install.

A coding agent runs vibedoctor_check on the files it changed and gets back a critical SQL injection with its file and line

In short

Agent Check is VibeDoctor inside your coding agent. Connect it once to Claude Code, Cursor or Codex. After a change, the agent sends the files it changed to vibedoctor_check and gets back every issue with its severity, line and a suggested fix, plus a list of what was and was not checked. It fixes the issues before you commit. There is no plugin to download and nothing to install.

How it works

  1. Connect VibeDoctor to your coding agent once (see Connect your agent).
  2. Your agent writes or edits code as usual.
  3. It sends the files it changed to vibedoctor_check.
  4. It gets back each issue with its severity, file, line and a suggested fix, then fixes them and checks again.

VibeDoctor asks the agent to run the check after every code change. Agents decide when to call a tool, so if yours skips one, ask it to check the files it changed.

What it checks

  • Security patterns: SQL injection, XSS, eval(), unprotected routes, CORS wildcards, insecure cookies and secrets in frontend code.
  • Secrets: API keys, passwords, tokens and private keys in the files the agent sends.
  • Deep analysis: rules that follow user input into databases, files and the shell.
  • Dependencies: when the agent sends package.json or another manifest, packages that do not exist and other dependency problems.

Every answer says what it checked and what it did not. A check looks only at the files the agent sends, so it does not run language linters, look up dependency CVEs or follow auth set up in other files. A full Vibe Check scan of your repository covers those.

Connect your agent

  • Claude Code and Codex: run one command and sign in. The connector page has the commands.
  • Cursor, VS Code and Windsurf: open MCP Connect in VibeDoctor, copy your project's link and add it to the editor's MCP settings (for Cursor, ~/.cursor/mcp.json).
  • Claude and ChatGPT: add VibeDoctor as a connector to ask about your scans and check code in chat.

That is the whole setup. There is no plugin, extension or script to install.

Every push scanned too

Agent Check covers the code your agent writes in a session. On the Watch, Guard and Shield plans, the VibeDoctor GitHub App also runs a full scan on every push to main or master, so a change that skipped the agent's check still gets scanned.

Your code

  • VibeDoctor receives only the files your agent sends for a check.
  • They are checked in a temporary folder that is deleted when the check ends.
  • VibeDoctor never sees your conversation with the agent or files it does not send.

The privacy policy has the details.

Limits

  • Free accounts get 50 calls per project per day. vibedoctor_status and vibedoctor_note do not count.
  • The count resets at midnight UTC.
  • Paid plans have no daily limit.
  • During your Free Week, a check returns every issue with its line and fix. After it, a check on the free plan shows how many issues it found, and a paid plan shows each one.

FAQ

Do I need to install anything?

No. There is no plugin or download. You add VibeDoctor to your agent once: one command in Claude Code or Codex, or one entry in Cursor's MCP settings.

Which coding agents work?

Claude Code, Codex, Cursor, VS Code and Windsurf. Claude and ChatGPT connect too, for questions about your scans and code checks in chat.

Does my agent check every change?

VibeDoctor asks it to run the check after every code change. Agents decide when to call a tool, so if yours skips one, ask it to check the files it changed.

How is this different from a full scan?

A check looks at the files your agent sends while it works. A full Vibe Check scans the whole repository and your live site, including language linters, dependency CVEs and auth set up across files.

Is Agent Check free?

Free accounts get 50 calls per project per day, and vibedoctor_status and vibedoctor_note do not count. During your Free Week each check shows every issue with its line and fix; after it, the free plan shows how many issues a check found. Paid plans show every issue, with no daily limit.

What happens to the code my agent sends?

It is checked in a temporary folder that is deleted when the check ends.

Want the long version? Read Claude Code Security Review: What Each Tool Catches (2026).

Try Agent Check free

Every new account starts with a Free Week: one project, one scan a day, and the full report, with no card required.

CREATE A FREE ACCOUNT →