YAML Policy Template for Development Agents

Development environments need policies that allow experimentation while still providing visibility. This template permits most actions but logs everything, making it easy to iterate on agent behavior while building a record of what the agent does.

version: "1.0"
name: "development-agent-policy"
description: "Permissive policy for development with full logging"

defaults:
  action: allow
  log: true
  notify: false

rules:
  # Block actions that could affect production systems
  - name: "block-production-access"
    match:
      tool: "database_query"
      parameters:
        connection:
          pattern: "prod|production"
    action: deny

  # Block external communications
  - name: "block-external-email"
    match:
      tool: "send_email"
    action: deny

  # Block financial transactions
  - name: "block-payments"
    match:
      tool: "process_payment"
    action: deny

  # Log but allow file system operations
  - name: "log-file-operations"
    match:
      tool: "file_*"
    action: allow
    log: true

  # Allow all read operations
  - name: "allow-reads"
    match:
      tool: "*_read"
    action: allow

  # Allow all search operations
  - name: "allow-search"
    match:
      tool: "*_search"
    action: allow

  # Require approval for anything touching credentials
  - name: "approve-credential-access"
    match:
      tool: "*"
      parameters:
        _any:
          pattern: "(password|secret|token|api_key)"
    action: approve
    approval:
      timeout: 600
      approvers: ["developer"]

Key differences from the production template:

Allow by default. Development agents can use any tool not explicitly blocked. This reduces friction during iteration and testing.

Hard blocks on production resources. Even in development, the policy prevents accidental access to production databases, email systems, and payment processors. These rules are non-negotiable regardless of environment.

Credential protection. Any tool call with parameters that look like credentials requires approval. This prevents accidental exposure even in permissive environments.

Full logging. Every action is logged. This creates a record that can be analyzed to build the production policy. Review the development logs to understand which tools the agent uses, with what parameters, and how frequently.

Use development logs to refine your production policy. The patterns you observe during development inform the allowlists and constraints you enforce in production.