Skip to content

Latest commit

 

History

History
111 lines (89 loc) · 2.6 KB

README.md

File metadata and controls

111 lines (89 loc) · 2.6 KB

Active Agent: Action Prompt

Action Prompt provides a structured way to create and manage prompts for AI interactions. It enables composing messages, handling actions/tools, and managing conversation context through several key components.

Action Prompt manages the overall prompt structure including:

  • Messages list with system/user/assistant roles
  • Action/tool definitions
  • Headers and context tracking
  • Content type and encoding
  • Multipart message handling
prompt = ActionPrompt::Prompt.new(
  instructions: "System guidance",
  message: "User input",
  actions: [tool_definition, action_schema],
  context: messages
)

ActionPrompt::Message

Represents individual messages with:

Content and Role

message = ActionPrompt::Message.new(
  content: "Search for a hotel",
  role: :user
)
  • Content stores the actual message text
  • Role defines the message sender type (system/user/assistant/function/tool)
  • Messages form interactions between roles in a Context

Action Requests and Responses

message = ActionPrompt::Message.new(
  content: "Search for a hotel",
  role: :tool,
  requested_actions: [{name: "search", params: {query: "hotel"}}]
)
  • Tracks if message requests actions/tools
  • Maintains list of requested_actions
  • Action responses include function call results
  • Handles tool execution state

Content Type and Encoding

  • Default content_type is "text/plain"
  • Supports multiple content types for rich messages
  • Default charset is UTF-8
  • Handles content encoding/decoding

Role Validation

  • Enforces valid roles via VALID_ROLES constant
  • Validates on message creation
  • Raises ArgumentError for invalid roles
  • Supported roles: system, assistant, user, tool, function
message = ActionPrompt::Message.new(
  content: "Hello",
  role: :user,
  content_type: "text/plain",
  charset: "UTF-8"
)

ActionPrompt::Action

Defines callable tools/functions:

  • Name and parameters schema
  • Validation and type checking
  • Response handling
  • Action execution

Usage with Base Agents

class MyAgent < ActiveAgent::Base
  # Define available actions
  def action_schemas
    [
      {name: "search", params: {query: :string}}
    ]
  end

  # Handle action responses
  def perform_action(action) 
    case action.name
    when "search"
      search_service.query(action.params[:query])
    end
  end
end

The Base agent integrates with ActionPrompt to:

  1. Create prompts with context
  2. Register available actions
  3. Process action requests
  4. Handle responses and update context
  5. Manage the conversation flow

See Base.rb for full agent integration details.