Confident Tracing Skill

Overview

The confident-tracing Agent Skill teaches your coding agent how to instrument an AI application with the Python or TypeScript confident-trace SDK so every model call, retrieval, tool call, and agent step appears span by span in Confident AI.

Its scope is deliberately narrow: producing well-formed traces. The skill detects the language, framework, model provider, agent SDK, gateway, bundler, and existing OpenTelemetry setup; prefers a supported integration; falls back to custom @span, span(), or withSpan() instrumentation where needed; and adds useful trace context. Attaching metrics and running evals belongs to the deepeval skill.

When It Triggers

The skill activates on prompts like:

Instrument this app with Confident Trace.
Add automatic tracing to my LangGraph agent.
Add @span around the custom retriever in this Python RAG pipeline.
Make my TypeScript agent's OpenAI calls appear in Confident AI.

Installation

npx skills add confident-ai/confident-trace --skill "confident-tracing"

/plugin marketplace add confident-ai/confident-trace
/plugin install confident-trace@confident-trace-plugins
/reload-plugins

git clone https://github.com/confident-ai/confident-trace
cp -r confident-trace/skills/confident-tracing .claude/skills/

Prerequisites

• Python 3.10+ or Node.js 22+
• confident-trace installed in the application
• A project-scoped CONFIDENT_API_KEY for export to Confident AI

What Changes in Your Codebase

1. Ask for tracing

   The agent detects the language and AI stack, reads the current confident-trace integration documentation, and chooses automatic instrumentation whenever a supported integration exists.

   Instrument this agent with confident-trace and send its traces to Confident AI.

2. Initialize the SDK

   In Python, the agent installs confident-trace and calls init() once before provider or framework calls:

   import confident_trace as ct
   
   ct.init()

   In TypeScript, automatic instrumentation requires both init() in the entry file and the Node registration preload:

   import { init } from "confident-trace";
   
   const tracing = init();

   node --import confident-trace/register dist/index.js

3. Add custom spans where integrations cannot

   The agent adds custom spans only around application-owned boundaries or unsupported components. It uses one of the five supported span types: agent, llm, retriever, tool, or custom.

   Python example

   import confident_trace as ct
   
   @ct.span(type="retriever")
   def retrieve(query: str) -> list[str]:
       documents = search(query)
       ct.update_span(input=query, retrieval_context=documents)
       return documents

   TypeScript example

   import { span, updateSpan } from "confident-trace";
   
   const retrieve = span(
     { name: "retrieve", type: "retriever" },
     async (query: string) => {
       const documents = await search(query);
       updateSpan({ input: query, retrievalContext: documents });
       return documents;
     },
   );

4. Add trace and conversation context

   The agent uses update_trace() or updateTrace() for trace input, output, tags, metadata, user ID, thread ID, turn ID, and environment. It uses turn() when each conversation turn must start a separate trace while remaining associated with the same thread.

   with ct.turn(thread_id="chat-42", turn_id="2", input=user_input):
       answer = run_agent(user_input)
       ct.update_trace(output=answer)

5. Verify traces

   The agent finishes active work and streams before flushing or shutting down, then verifies the trace hierarchy in Confident AI:

   export CONFIDENT_API_KEY="confident_us_proj_..."
   python main.py

FAQs

Next Steps

Last updated on September 11, 2026