Confident OpenTelemetry Skill

Overview

The confident-otel Agent Skill teaches your coding agent how to export raw OpenTelemetry traces from an AI application to Confident AI without the confident-trace package. It works with any OpenTelemetry SDK in any language because the contract is the OTLP exporter endpoint plus the confident.* attributes on each span.

The skill points an OTLP/HTTP traces exporter at the correct Confident AI region, adds the x-confident-api-key header, preserves native OpenTelemetry trace context, and sets confident.span.* and confident.trace.* fields. Parentage, trace IDs, sampling, status, resources, links, and propagation remain native OpenTelemetry concerns.

Only confident_eu_... API keys use the EU endpoint.

When It Triggers

The skill activates on prompts like:

Send our OpenTelemetry AI traces to Confident AI.
Wire our OTel Collector to export agent spans to Confident AI.
Set confident.span.type on the LLM spans in our TypeScript service.
Which direct OTLP endpoint should our EU deployment use?

Installation

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

/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-otel .claude/skills/

Prerequisites

• A project-scoped CONFIDENT_API_KEY
• An OpenTelemetry SDK for the application's language
• For Python, opentelemetry-sdk and opentelemetry-exporter-otlp-proto-http

What Changes in Your Codebase

1. Inspect the existing OpenTelemetry setup

   The agent checks for a TracerProvider, span processors, exporters, Collector configuration, and APM instrumentation. It preserves the application's provider and prefers repointing an existing exporter over adding a duplicate pipeline.

   Send our OpenTelemetry AI traces to Confident AI.

2. Configure direct OTLP export

   The agent selects the region endpoint and configures the x-confident-api-key header. Standard OpenTelemetry environment variables can configure the direct exporter without a package-specific SDK:

   export OTEL_EXPORTER_OTLP_ENDPOINT="https://otel.confident-ai.com"
   export OTEL_EXPORTER_OTLP_HEADERS="x-confident-api-key=<CONFIDENT_API_KEY>"

   Python example

   import os
   
   from opentelemetry import trace
   from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
   from opentelemetry.sdk.trace import TracerProvider
   from opentelemetry.sdk.trace.export import BatchSpanProcessor
   
   provider = TracerProvider()
   provider.add_span_processor(
       BatchSpanProcessor(
           OTLPSpanExporter(
               endpoint="https://otel.confident-ai.com/v1/traces",
               headers={"x-confident-api-key": os.environ["CONFIDENT_API_KEY"]},
           )
       )
   )
   trace.set_tracer_provider(provider)

3. Set AI span and trace fields

   The agent sets confident.span.* fields on AI components and confident.trace.* fields for the whole trace. It JSON-encodes objects and metadata, uses native string arrays for lists, and preserves native OpenTelemetry parent-child context.

   Raw OpenTelemetry example

   tracer = trace.get_tracer(__name__)
   
   with tracer.start_as_current_span("support-agent") as root:
       root.set_attribute("confident.span.type", "agent")
       root.set_attribute("confident.trace.name", "support-chat")
   
       with tracer.start_as_current_span("chat-completion") as llm:
           llm.set_attribute("confident.span.type", "llm")
           llm.set_attribute("confident.llm.model", "gpt-4o")

4. Keep non-AI spans out

   If the process also emits HTTP, database, cache, or infrastructure spans, the agent uses a dedicated AI provider or filters the Confident AI-bound processor or exporter. It preserves parentage when removing intermediate spans.

5. Flush and verify

   The component that owns the provider owns shutdown. The agent finishes active work and streams, flushes or shuts down the provider, and verifies the AI trace hierarchy in Confident AI.

Attribute Rules

• Raw span types are llm, tool, agent, retriever, and base.
• Set confident.span.type explicitly when known.
• Put trace-wide input, output, tags, metadata, environment, user ID, thread ID, turn ID, and metric collection on confident.trace.*.
• Put component input, output, metadata, retrieval context, expected output, tools, and metric collection on confident.span.*.
• Use native OpenTelemetry status and exception recording for errors.
• Existing gen_ai.* attributes can provide fallbacks for model, token counts, tool name, and basic span-type inference.

FAQs

Next Steps

Last updated on September 11, 2026