Exported and Public Playbooks

Exported Playbooks

Exported playbooks are a powerful feature of Playbooks AI that allows an agent to execute playbooks defined by another agent within its own execution context as if they were implemented locally within the importing agent.

Playbooks AI's exported playbooks capability is significantly more powerful than MCP's prompts because it enables true capability sharing and execution between agents, not just structured guidance. While MCP prompts are valuable for standardizing LLM interactions, Playbooks AI enables building extensible agent ecosystems with genuine code reuse and modular architectures.

Public Playbooks

When an agent marks a playbook as public, another agent can call that playbook remotely.

Public playbooks are similar to MCP's tools. When a client calls a tool, it is executed on the MCP server. Similarly, when an agent calls a public playbook on another agent, it is executed on the called agent's server.

How Exported Playbooks Work

When an agent exports a playbook, another agent can import implementation of that playbook and then execute that playbook within its execution context.

Exporting Playbooks

This feature under active development and not available yet.

Exported playbooks are marked using the export keyword:

# AccountManagementAgent

```python
@playbook(export=True)
async def CloseAccount($user_id) -> str:
    """Close an account for a user."""
    # ...
```

## export: ProvisionAccount($user_id, $service_tier)
### Steps
- ...

In this example, the AccountManagementAgent exports the CloseAccount and ProvisionAccount playbooks.

Implementations of both markdown and Python playbooks can be exported.

Importing Exported Playbook Implementations

To import a playbook's implementation from another remote agent, register that remote agent and then use the import statement:

# AccountManagementAgent(url="https://cloud.runplaybooks.ai/account-management")

# ServiceAgent
import CloseAccount, ProvisionAccount from AccountManagementAgent

This imports implementations of the CloseAccount Python playbook and the ProvisionAccount markdown playbook from the AccountManagementAgent agent.

Using Imported Playbooks

Once imported, you can use the playbook just like any other playbook defined within your agent:

## CreateAccount
### Steps
- Validate user information
- Create user record in database
- ProvisionAccount($user_id, $service_tier)
- Send welcome email to user

The imported ProvisionAccount playbook is called as if it were a local playbook.

How Public Playbooks Work

This feature under active development and not available yet.

When agent A marks a playbook as public, another agent B can call that playbook remotely on a running instance of agent A.

Marking Playbooks as Public

To mark a playbook as public, use the public keyword:

# AuthenticationAgent
This agent handles user authentication.

## public: VerifyCredentials($username, $password)
### Steps
- Check credentials against secure store
- Return authentication result and user details if valid

Both markdown and Python playbooks can be marked as public.

Calling Public Playbooks

To call a public playbook, first register the remote agent and then call the playbook as a method on that agent:

# AuthenticationAgent(url="https://cloud.runplaybooks.ai/authentication")

# APIGatewayAgent

## HandleRequest
### Steps
- $auth_result = AuthenticationAgent.VerifyCredentials($request.auth.username, $request.auth.password)
- ...

Here, the APIGatewayAgent directly calls the VerifyCredentials playbook on the AuthenticationAgent instance running at https://cloud.runplaybooks.ai/authentication. The playbook is executed on the remote agent's server and the result is returned to the calling agent.

Differences Between Exported and Public Playbooks

Exported playbooks differ from public playbooks in several important ways:

	Exported Playbooks	Public Playbooks
Example	import CloseAccount from AccountManagementAgent and then CloseAccount($user_id)	# PaymentProcessingAgent("https://cloud.runplaybooks.ai/ppa") and then PaymentProcessingAgent.ProcessPayment($amount, $payment_method)
Execution Context	Local execution in importing agent's context	Remote procedure call on the remote agent instance
State Access	Can access local agent's state	Cannot access local agent's state

Security Considerations

Be careful! Importing playbooks raise several security considerations.

1. Code Injection: An agent that imports playbooks from another agent must trust that agent. Importing playbook implementations from untrusted agents and executing them can lead to security vulnerabilities like code injection attacks.

2. Version Control: Changes to exported playbooks may affect importing agents, requiring careful version management.

Example: Agent Ecosystem

Let's say that we have an ecosystem of three agents.

1. DatabaseAgent

# DatabaseAgent
This agent provides recipes for various database related tasks.

## export: FindTable($query, $database)
### Steps
- List all tables in the $database
- Find the table that is most likely to contain data to answer the $query
- Return the table name

Let's say that DatabaseAgent is available at the URL https://cloud.runplaybooks.ai/database.agent. It exports the FindTable playbook. It is a generic procedure for finding a database table.

2. AuthenticationAgent

# AuthenticationAgent
This agent handles user authentication.

## public: VerifyCredentials($username, $password)
### Steps
- Check credentials against secure store
- Return authentication result and user details if valid

Let's say that an instance of the AuthenticationAgent is running at the URL https://cloud.runplaybooks.ai/authentication.agent. The public VerifyCredentials playbook requires access to the secure store within ACME Corp's infrastructure.

3. APIGatewayAgent

APIGatewayAgent uses the above two agents. It first registers those two agents, specifying their URLs. Then it imports the FindTable playbook from the DatabaseAgent. It then remotely calls the VerifyCredentials playbook on the AuthenticationAgent instance. Finally, it locally executes the FindTable playbook.

# DatabaseAgent(url="https://cloud.runplaybooks.ai/database.agent")

# AuthenticationAgent(url="https://cloud.runplaybooks.ai/authentication.agent")

# APIGatewayAgent
import FindTable from DatabaseAgent

## HandleRequest
### Triggers
- When an API request is received
### Steps
- $auth_result = AuthenticationAgent.VerifyCredentials($request.auth.username, $request.auth.password)
- If $auth_result.success is true
  - $table_name = FindTable($request.query, $database)
  - Return $table_name
- Otherwise
  - Return authentication error

Related Topics

• Multi-Agent Programming - More about setting up and using multiple agents
• Python Playbooks - How to create playbooks using Python
• Markdown Playbooks - Standard playbook structure