agentlint

AgentLint

Catch risky actions while your coding agent works.

AgentLint checks proposed file changes and shell commands locally. It blocks detected secret writes and plain force-pushes to main or master, warns about risky operations, and gives your agent feedback it can act on.

Quickstart · Documentation · Integration coverage · Rules

*Illustration with a supported pre-tool hook and standard policy. AgentLint evaluates the revised proposal again; it does not rewrite your code itself.*

See the difference

A coding agent proposes putting a credential in config.py. The no-secrets rule denies the write and returns a reason:

The agent can replace the literal with an environment-variable lookup:

The credential finding is resolved. The revised action is checked against the active policy, and any other findings still apply.

https://github.com/user-attachments/assets/ae9a6b72-8bf7-4b73-abe3-7f0f3b483d3b

Watch the 53-second product overview. The demonstration evaluates synthetic proposed actions as data; it does not execute dangerous commands.

Get started

Requires Python 3.11+ and pipx or uv.

Use your agent's name for the last command:

Prefer pipx? Use pipx install agentlint instead. Setup adds hooks to your agent's settings and creates agentlint.yml if it does not exist, using packs detected from the project.

Make one tool call in your agent, then run:

Look for a recent observed hook call from this project. A configuration file alone does not prove the hooks are running. If nothing is observed, run agentlint doctor and follow troubleshooting.

The five-minute quickstart completes the first check and correction. Claude Code users can alternatively install the marketplace plugin; use the plugin or native setup to avoid duplicate hooks.

What blocks, what warns

With…