Skip to content
PromptKit Logo PromptKit

Use Tool Bridge

The Tool Bridge converts an A2A agent’s skills into ToolDescriptor entries so they can be used through the standard tool registry — enabling LLMs to call remote agents as if they were local tools.

Overview

Section titled “Overview”
flowchart LR
    D["client.Discover()"] --> S["Agent Skills"]
    S --> TD["ToolDescriptors"]
    TD --> R["Tool Registry"]
    R --> LLM["LLM tool calling"]


    style D fill:#e3f2fd
    style R fill:#e8f5e9
  1. The bridge calls Discover() to fetch the agent card
  2. Each skill is converted to a ToolDescriptor
  3. Descriptors are registered in the tool registry
  4. The LLM can invoke them during tool-calling rounds

Register an A2A Agent as a Tool

Section titled “Register an A2A Agent as a Tool”
package main


import (
    "context"
    "fmt"
    "log"


    "github.com/AltairaLabs/PromptKit/runtime/a2a"
    "github.com/AltairaLabs/PromptKit/sdk"
)


func main() {
    ctx := context.Background()


    // Point a client at the remote agent.
    client := a2a.NewClient("http://localhost:9999")


    // Create the bridge and register the agent's skills.
    bridge := a2a.NewToolBridge(client)
    tools, err := bridge.RegisterAgent(ctx)
    if err != nil {
        log.Fatal(err)
    }


    for _, t := range tools {
        fmt.Printf("Registered: %s — %s\n", t.Name, t.Description)
    }


    // Use the bridge with the SDK.
    conv, err := sdk.Open("assistant.pack.json", "chat",
        sdk.WithA2ATools(bridge),
    )
    if err != nil {
        log.Fatal(err)
    }
    defer conv.Close()


    resp, err := conv.Send(ctx, "Search for papers about quantum computing")
    if err != nil {
        log.Fatal(err)
    }
    fmt.Println(resp.Text())
}

Naming Convention

Section titled “Naming Convention”

Tool names follow the pattern:

a2a__{agentName}__{skillId}

Both the agent name and skill ID are sanitized: lowercased, with non-alphanumeric runs replaced by underscores. For example:

Agent NameSkill IDTool Name
Research Agentsearch_papersa2a__research_agent__search_papers
Code Helperlint-codea2a__code_helper__lint_code

This naming is important for allowed_tools in prompt configs — you reference the tool by its generated name.

Use with Prompt Config

Section titled “Use with Prompt Config”

To enable A2A tools in a prompt config, list them in allowed_tools:

apiVersion: promptkit.altairalabs.ai/v1alpha1
kind: PromptConfig
spec:
  allowed_tools:
    - a2a__research_agent__search_papers
  system_template: |
    You are a helpful assistant with access to research tools.
    When asked to search for papers, use the available tools.

The tool calling pipeline automatically invokes the A2A agent when the LLM decides to call the tool.

Multi-Agent Composition

Section titled “Multi-Agent Composition”

Register multiple agents on the same bridge. Each RegisterAgent call appends to the bridge’s internal tool list:

bridge := a2a.NewToolBridge(researchClient)
bridge.RegisterAgent(ctx) // adds research agent tools


// Point the bridge at another agent.
bridge2 := a2a.NewToolBridge(translationClient)
bridge2.RegisterAgent(ctx)

Or discover from multiple agents and combine:

bridge := a2a.NewToolBridge(client1)
bridge.RegisterAgent(ctx) // agent 1 tools


// The bridge only works with one client, so create a second bridge.
bridge2 := a2a.NewToolBridge(client2)
bridge2.RegisterAgent(ctx) // agent 2 tools


// Use both with SDK.
conv, _ := sdk.Open("assistant.pack.json", "chat",
    sdk.WithA2ATools(bridge),
    sdk.WithA2ATools(bridge2),
)

Input/Output Schema Generation

Section titled “Input/Output Schema Generation”

The bridge automatically generates JSON Schemas for each tool based on the skill’s input/output modes:

  • Text — always includes a query field (required)
  • Image input modes (image/*) — adds image_url and image_data fields
  • Audio input modes (audio/*) — adds audio_data field
  • Media output modes — adds media_url and media_type fields

Next Steps

Section titled “Next Steps”