Agent Handler Skill

Overview

The confident-agent-handler Agent Skill teaches your coding agent how to connect an AI app that has no HTTP endpoint to Confident AI through Handler Mode. The agent finds your pipeline in your code, writes a small handler function that calls it, gives you the command to run the Confident Agent beside it, then creates an Agent Handler AI connection and pings it until it works.

The handler is the only file added to your codebase. Your pipeline runs where it already does: the agent dials out to Confident AI, receives each input over that outbound connection, and calls your handler in-process, so nothing is inbound and your code never leaves your network.

When It Triggers

The skill activates on prompts like:

Connect our summarization pipeline to Confident AI. It has no API.
Set up handler mode so Confident AI can red team our internal agent.
Write a Confident Agent handler for the notebook in research/.
Our Agent Handler connection fails its ping, figure out why and fix it.

Installation

npx skills add confident-ai/confident-skills --skill "confident-agent-handler"

/plugin marketplace add confident-ai/confident-skills
/plugin install confident-eval-setup@confident-skills
/reload-plugins

git clone https://github.com/confident-ai/confident-skills
cp -r confident-skills/plugins/confident-eval-setup/skills/confident-agent-handler .claude/skills/

Prerequisites

• An Enterprise plan with a project API key for the agent, in CONFIDENT_API_KEY. The agent never sees it: it writes the run command with a placeholder for you to fill in.
• Outbound access on port 443 from the machine running the agent, and a runtime where your pipeline is importable.
• The Confident AI MCP server connected to your coding agent, so it can create and ping the connection. Without it, the agent walks you through those two steps in the dashboard instead.

What Happens in Your Project

1. Ask for a connection

   The agent calls list_projects to find the project to work in, then list_ai_connections, so it fixes an existing Agent Handler connection instead of creating a duplicate.

   Connect our summarization pipeline to Confident AI. It has no API.

2. Write the handler

   The agent finds the function that turns an input into your app's answer and writes a handler beside it in the same language (Python, TypeScript, Go, Java, or Rust). The handler imports your pipeline, calls it, and returns a Response with the output, plus the retrieval context and tools called when your pipeline has them.

   See what the agent writes

   from confident_agent import handler, Request, Response
   from my_company.pipeline import summarize
   
   @handler
   def summarize_handler(request: Request) -> Response:
       return Response(
           output=summarize(request.input, request.context),
       )

3. Run the agent

   The agent installs confident-agent into your pipeline's own environment and gives you the command to start it, with your project API key as a placeholder:

   CONFIDENT_API_KEY=<your-api-key> confident-agent --handler confident_handler.py

   There is no URL or port to configure. Once it connects, the connection status shows Agent connected.

4. Create the connection and ping it

   The agent creates the connection with create_ai_connection and type: "AGENT_HANDLER". There is no endpoint, payload, or output key path to set, because Response fields are mapped automatically.

   See what the agent sends

   {
     "name": "Summarization pipeline",
     "type": "AGENT_HANDLER"
   }

   It then calls ping_ai_connection, which sends one sample input through the tunnel and runs your handler. If the ping fails, the agent reads the error, fixes the handler or how the agent runs, and pings again until it passes.

   Our Agent Handler connection fails its ping, figure out why and fix it.

FAQs

Next Steps

Last updated on November 6, 2026