Would Verfy let your agent run this?
Type a shell command, as an AI coding agent would run it. The same rules Verfy's hook uses decide here, in your browser: allow, ask you first, or deny. Nothing is run and nothing is sent.
Examples:
How it decides
- The rules are fixed patterns, written in the open: no model decides allow or deny, so the same command always gets the same answer.
- Commands are read the way a shell reads them: pipes,
&&, quotes, variables,bash -c,find -exec,xargsand wrappers such assudoorenv. - The policy here is Verfy's default, for a home folder
/home/youand a project in/home/you/project. Yours can be stricter or allow more.
What it cannot do
- It reads the command, not what a script does once it runs. A script the agent writes and runs is held by the agent's own sandbox;
verfy posture --fixturns it on. - Pattern rules can be phrased around. If you find a command that should be stopped and is not, report it.
- Windows commands are checked with patterns on the text, which is weaker than the shell parsing for bash.
Put the check in front of your agent
One command sets it up for every coding agent on your computer: Claude Code, Codex CLI, Gemini CLI, Cursor, GitHub Copilot and Devin.
$ pip install verfy
$ verfy setup --write