Skip to main content

Policy framework for Claude Code and Cursor hooks

Project description

AI Agent Policies

PyPI

Policies turn your Cursor Rules or CLAUDE.md into hard guardrails which an AI Agent cannot simply ignore, or forget. They handle what to do when an agent wants to make a decision, along with other hooks-supported events. Policies can yield both decisions and guidance.

This framework supports Claude Code. Support for Cursor is in beta.

Why Policies

Policies are significantly stronger than rules. Rule files can be forgotten or ignored completely, whereas policies are unavoidable:

def rudimentary_terraform_rule(input_data: ToolUseEvent):
    if not input_data.tool_is_bash:
        return

    command = input_data.command.strip()

    if re.match(r'^terraform\s+apply(?:\s|$)', command):
        yield PolicyDecision(action=PolicyAction.DENY, reason="terraform apply is not allowed. Use `terraform plan` instead.")

    if re.match(r'^terraform\s+(fmt|plan)(?:\s|$)', command):
        yield PolicyDecision(action=PolicyAction.ALLOW)

Aside from denying and allowing automatically, policies can also provide guidance when it seems agents are going off track or missing standards:

def rudimentary_guidance_for_python(input_data: ToolUseEvent):
    if not input_data.tool_is_bash:
        return

    if re.match(r'^python\s+test_', input_data.command):
        yield PolicyGuidance(content="Consider using pytest instead of running test files directly")

Be aware that tool use with bash can contain control operators and separators, moreover, some commands allow execution of others: Consider for example find with -exec. Be careful what you whitelist, and how you parse Bash. At DevLeaps we also have an internal policy set, which is not included in this project. To create your own, refer to the example server which contains a rudimentary bash middleware to demonstrate what is possible with the framework.

Examples

Real World Examples

Policies can prevent Bash dangerous bash commands completely:

Policies can provide guidance to agents when they go off track:

Policies can automatically allow safe commands without interrupting developers:

Usage

Have a look at the example server as a starting point!

It contains:

  • A complete example server with policies and middleware.
  • Rudimentary policies on how to automatically deny, allow and provide guidance.
  • Rudimentary middleware demonstrating how multi-command tool use could be handled.

To run the example server:

devleaps-policy-example-server

This starts a minimal server with just these example policies.

Architecture

graph TB
    subgraph "Developer Machine"
      Editor[Claude Code / Cursor]
        Client[devleaps-policy-client]
    end

    subgraph "Policy Server"
        Server[HTTP API]
        Policies[Your policies<br/>kubectl, terraform, git, python, etc.]
    end

    Editor -->|Hooks| Client
    Client --> Server
    Server -->|Events| Policies
    Policies -->|Decision and Guidance| Server
    Server --> Client
    Client -->|Decision and Guidance| Editor

Quick Start

Installation

Update your local profile with;

# Add the bin directory to $PATH
export PATH="$PATH:/path/to/agent-policies/bin/"

Running an Example Server

devleaps-policy-example-server

The example server runs on port 8338 by default and serves endpoints for both Claude Code and Cursor.

Configure Claude Code

Add devleaps-policy-client to your Claude Code hooks configuration in ~/.claude/settings.json:

Click to expand Claude Code configuration
{
  "hooks": {
    "PreToolUse": [
      {
        "hooks": [
          {
            "matcher": "*",
            "type": "command",
            "command": "devleaps-policy-client claude-code"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "hooks": [
          {
            "matcher": "*",
            "type": "command",
            "command": "devleaps-policy-client claude-code"
          }
        ]
      }
    ],
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "matcher": "*",
            "type": "command",
            "command": "devleaps-policy-client claude-code"
          }
        ]
      }
    ],
    "Stop": [
      {
        "hooks": [
          {
            "matcher": "*",
            "type": "command",
            "command": "devleaps-policy-client claude-code"
          }
        ]
      }
    ],
    "SubagentStop": [
      {
        "hooks": [
          {
            "matcher": "*",
            "type": "command",
            "command": "devleaps-policy-client claude-code"
          }
        ]
      }
    ],
    "Notification": [
      {
        "hooks": [
          {
            "matcher": "*",
            "type": "command",
            "command": "devleaps-policy-client claude-code"
          }
        ]
      }
    ],
    "PreCompact": [
      {
        "hooks": [
          {
            "matcher": "*",
            "type": "command",
            "command": "devleaps-policy-client claude-code"
          }
        ]
      }
    ],
    "SessionStart": [
      {
        "hooks": [
          {
            "matcher": "*",
            "type": "command",
            "command": "devleaps-policy-client claude-code"
          }
        ]
      }
    ],
    "SessionEnd": [
      {
        "hooks": [
          {
            "matcher": "*",
            "type": "command",
            "command": "devleaps-policy-client claude-code"
          }
        ]
      }
    ]
  }
}

Configure Cursor

Create or edit ~/.cursor/hooks.json:

Click to expand Cursor configuration
{
  "version": 1,
  "hooks": {
    "beforeShellExecution": [
      { "command": "devleaps-policy-client cursor" }
    ],
    "beforeMCPExecution": [
      { "command": "devleaps-policy-client cursor" }
    ],
    "afterFileEdit": [
      { "command": "devleaps-policy-client cursor" }
    ],
    "beforeReadFile": [
      { "command": "devleaps-policy-client cursor" }
    ],
    "beforeSubmitPrompt": [
      { "command": "devleaps-policy-client cursor" }
    ],
    "stop": [
      { "command": "devleaps-policy-client cursor" }
    ]
  }
}

The devleaps-policy-client cursor command will forward hook events to the policy server running on localhost:8338.

Development

This project is built with uv.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

devleaps_agent_policies-1.0.1.tar.gz (44.9 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

devleaps_agent_policies-1.0.1-py3-none-any.whl (32.1 kB view details)

Uploaded Python 3

File details

Details for the file devleaps_agent_policies-1.0.1.tar.gz.

File metadata

File hashes

Hashes for devleaps_agent_policies-1.0.1.tar.gz
Algorithm Hash digest
SHA256 6403fc3f827d3f3b8d6f7b099698694e8a3577d02142e9426292a5b65756025e
MD5 393d56fc62f90b1c15aa54780323cfc4
BLAKE2b-256 37e4d14fafebf2ce8f95eb6aabd75cfc1163daee412b1c71f64a5f3bbe51fed9

See more details on using hashes here.

File details

Details for the file devleaps_agent_policies-1.0.1-py3-none-any.whl.

File metadata

File hashes

Hashes for devleaps_agent_policies-1.0.1-py3-none-any.whl
Algorithm Hash digest
SHA256 6ea252e8d86eef8f4227c255af853c223014fb74edbb1672e49f9893cfb98bb8
MD5 edcc38b67dea617d185a3a224f38ce47
BLAKE2b-256 61c807f63d13254fdb7b38143eb42770b9b511028671fd51ef766311406eec89

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page