arize-instrumentation

SKILL.md

Arize Instrumentation Skill

Use this skill when the user wants to add Arize AX tracing to their application. Follow the two-phase, agent-assisted flow from the Agent-Assisted Tracing Setup and the Arize AX Tracing — Agent Setup Prompt.

Quick start (for the user)

If the user asks you to "set up tracing" or "instrument my app with Arize", you can start with:

Follow the instructions from https://arize.com/docs/PROMPT.md and ask me questions as needed.

Then execute the two phases below.

Core principles

Prefer inspection over mutation — understand the codebase before changing it.

Do not change business logic — tracing is purely additive.

Use auto-instrumentation where available — add manual spans only for custom logic not covered by integrations.

Follow existing code style and project conventions.

Keep output concise and production-focused — do not generate extra documentation or summary files.

NEVER embed literal credential values in generated code — always reference environment variables (e.g., os.environ["ARIZE_API_KEY"], process.env.ARIZE_API_KEY). This includes API keys, space IDs, and any other secrets. The user sets these in their own environment; the agent must never output raw secret values.

Phase 0: Environment preflight

Before changing code:

Confirm the repo/service scope is clear. For monorepos, do not assume the whole repo should be instrumented.

Identify the local runtime surface you will need for verification:

package manager and app start command

whether the app is long-running, server-based, or a short-lived CLI/script

whether ax will be needed for post-change verification

Do NOT proactively check ax installation or version. If ax is needed for verification later, just run it when the time comes. If it fails, see references/ax-profiles.md.

Never silently replace a user-provided space ID, project name, or project ID. If the CLI, collector, and user input disagree, surface that mismatch as a concrete blocker.

When you must ask the user first

If monorepo scope, service entrypoint, or target app is still unclear after quick inspection — or you would otherwise open with a bare list of questions — use this opening pattern:

Acknowledge the skill, e.g.: I found the arize-instrumentation skill in this repo (you may add skills/arize-instrumentation/SKILL.md if helpful).

Then a clear pause line, e.g.: A few clarifying questions before I invoke it:

Ask minimal numbered or short bullet questions — only what blocks Phase 1 or Phase 2.

Phase 1: Analysis (read-only)

Do not write any code or create any files during this phase.

Steps

Check dependency manifests to detect stack:

Python: pyproject.toml, requirements.txt, setup.py, Pipfile

TypeScript/JavaScript: package.json

Java: pom.xml, build.gradle, build.gradle.kts

Go: go.mod

Scan import statements in source files to confirm what is actually used.

Check for existing tracing/OTel — look for TracerProvider, register(), opentelemetry imports, ARIZE_*, OTEL_*, OTLP_* env vars, or other observability config (Datadog, Honeycomb, etc.).

Identify scope — for monorepos or multi-service projects, ask which service(s) to instrument.

What to identify

Item
Examples

Language
Python, TypeScript/JavaScript, Java, Go

Package manager
pip/poetry/uv, npm/pnpm/yarn, maven/gradle, go modules

LLM providers
OpenAI, Anthropic, LiteLLM, Bedrock, etc.

Frameworks
LangChain, LangGraph, LlamaIndex, Vercel AI SDK, Mastra, etc.

Existing tracing
Any OTel or vendor setup

Tool/function use
LLM tool use, function calling, or custom tools the app executes (e.g. in an agent loop)

Key rule: When a framework is detected alongside an LLM provider, inspect the framework-specific tracing docs first and prefer the framework-native integration path when it already captures the model and tool spans you need. Add separate provider instrumentation only when the framework docs require it or when the framework-native integration leaves obvious gaps. If the app runs tools and the framework integration does not emit tool spans, add manual TOOL spans so each invocation appears with input/output (see references/manual-spans.md).

Phase 1 output

Return a concise summary:

Detected language, package manager, providers, frameworks

Proposed integration list (from the routing table in the docs)

Any existing OTel/tracing that needs consideration

If monorepo: which service(s) you propose to instrument

If the app uses LLM tool use / function calling: note that you will add manual CHAIN + TOOL spans so each tool call appears in the trace with input/output (avoids sparse traces).

If the user explicitly asked you to instrument the app now, and the target service is already clear, present the Phase 1 summary briefly and continue directly to Phase 2. If scope is ambiguous, or the user asked for analysis first, stop and wait for confirmation.

Integration routing and docs

Use the Agent Setup Prompt routing table to map detected signals to integration docs and fetch the matched pages for exact installation steps and code snippets. Use llms.txt as a fallback for doc discovery.

See references/integration-routing.md for the full list of supported integrations by language and platform.

Phase 2: Implementation

Proceed only after the user confirms the Phase 1 analysis.

Steps

Fetch integration docs — Read the matched doc URLs and follow their installation and instrumentation steps.

Install packages using the detected package manager before writing code:

Python: pip install arize-otel plus openinference-instrumentation-{name} (hyphens in package name; underscores in import, e.g. openinference.instrumentation.llama_index).

TypeScript/JavaScript: @opentelemetry/sdk-trace-node plus the relevant @arizeai/openinference-* package.

Java: OpenTelemetry SDK plus openinference-instrumentation-* in pom.xml or build.gradle.

Go: Use arize-otel-go for tracer setup, plus a per-provider instrumentor when one exists. Install:
go get github.com/Arize-ai/arize-otel-go
go get github.com/Arize-ai/openinference/go/openinference-semantic-conventions
go get github.com/Arize-ai/openinference/go/openinference-instrumentation
# Plus exactly one of these, matched to the detected client:
go get github.com/Arize-ai/openinference/go/openinference-instrumentation-openai-go        # official openai/openai-go SDK
go get github.com/Arize-ai/openinference/go/openinference-instrumentation-anthropic-sdk-go # anthropics/anthropic-sdk-go v1.43+

Wire the exporter with one call: arizeotel.Register(ctx, arizeotel.Options{ProjectName: "my-app"}) — defaults to otlp.arize.com (US), use arizeotel.EndpointArizeEurope for EU. It reads ARIZE_SPACE_ID / ARIZE_API_KEY / ARIZE_PROJECT_NAME / ARIZE_COLLECTOR_ENDPOINT from env when the matching Options fields are unset. Wire the OpenAI instrumentor by passing option.WithMiddleware(openaiotel.Middleware(otel.Tracer("my-app"))) to openai.NewClient(...) (alongside option.WithAPIKey(...)). Wire the Anthropic instrumentor by passing option.WithMiddleware(anthropicotel.Middleware(otel.Tracer("my-app"))) to anthropic.NewClient(...). Both instrumentors expose WithTraceConfig(instrumentation.TraceConfig{...}) for in-code overrides of the OPENINFERENCE_HIDE_* env-driven masking config. Module floor is Go 1.25 (the openinference Go modules require it; arize-otel-go itself is Go 1.23+).

Credentials — User needs an Arize API Key and Space ID. Check existing ax profiles for ARIZE_API_KEY and ARIZE_SPACE — never read .env files:

Run ax profiles show to check for an existing profile. Run ax profiles validate to verify an existing profile's credentials are still valid.

If no profile exists, guide the user to run ax profiles create which provides an interactive wizard that walks through API key and space setup. See CLI profiles docs for details.

OAuth alternative (v0.18.0+): Users can authenticate via browser-based OAuth PKCE instead of API keys by running ax auth login. Inform users of this option if they prefer not to manage API keys — do not run ax auth login yourself as it opens a browser.

If the user needs to find their API key manually, direct them to https://app.arize.com and to navigate to the settings page (do not use organization-specific URLs with placeholder IDs — they won't resolve for new users).

If credentials are not set, instruct the user to set them as environment variables — never embed raw values in generated code. All generated instrumentation code must reference os.environ["ARIZE_API_KEY"] / os.environ["ARIZE_SPACE"] (Python), process.env.ARIZE_API_KEY / process.env.ARIZE_SPACE (TypeScript/JavaScript), or os.Getenv("ARIZE_API_KEY") / os.Getenv("ARIZE_SPACE_ID") (Go — arize-otel-go reads ARIZE_SPACE_ID, not ARIZE_SPACE). With the recommended arizeotel.Register(ctx, arizeotel.Options{...}) flow, generated Go code does not need to call os.Getenv at all — Register reads both env vars when the matching Options fields are unset.

See references/ax-profiles.md for full profile setup and troubleshooting.

Centralized instrumentation — Create a single module (e.g. instrumentation.py, instrumentation.ts, instrumentation.go) and initialize tracing before any LLM client is created.

Existing OTel — If there is already a TracerProvider, add Arize as an additional exporter (e.g. BatchSpanProcessor with Arize OTLP). Do not replace existing setup unless the user asks.

Implementation rules

Use auto-instrumentation first; manual spans only when needed.

Prefer the repo's native integration surface before adding generic OpenTelemetry plumbing. If the framework ships an exporter or observability package, use that first unless there is a documented gap.

Fail gracefully if env vars are missing (warn, do not crash).

Import order: register tracer → attach instrumentors → then create LLM clients.

Project name attribute (required): Arize rejects spans with HTTP 500 if the project name is missing — service.name alone is not accepted. Set it as a resource attribute on the TracerProvider (recommended — one place, applies to all spans):

Python: register(project_name="my-app") handles it automatically (sets "openinference.project.name" on the resource). For routing spans to different projects, use set_routing_context(space_id=..., project_name=...) from arize.otel.

TypeScript: Arize accepts both "model_id" (shown in the official TS quickstart) and "openinference.project.name" via SEMRESATTRS_PROJECT_NAME from @arizeai/openinference-semantic-conventions (shown in the manual instrumentation docs) — both work.

Go: arizeotel.Register(ctx, arizeotel.Options{ProjectName: "my-app"}) handles this automatically (sets openinference.project.name and service.name on the resource). If you're wiring sdktrace.NewTracerProvider directly (multi-exporter, on-prem collector), pass attribute.String("openinference.project.name", "my-app") to resource.New(...) manually.

CLI/script apps — flush before exit: provider.shutdown() (TS) / provider.force_flush() then provider.shutdown() (Python) / tp.Shutdown(ctx) (Go) must be called before the process exits, otherwise async OTLP exports are dropped and no traces appear.

When the app has tool/function execution: add manual CHAIN + TOOL spans (see references/manual-spans.md) so the trace tree shows each tool call and its result — otherwise traces will look sparse (only LLM API spans, no tool input/output).

Verification

Treat instrumentation as complete only when all of the following are true:

The app still builds or typechecks after the tracing change.

The app starts successfully with the new tracing configuration.

You trigger at least one real request or run that should produce spans.

You either verify the resulting trace in Arize, or you provide a precise blocker that distinguishes app-side success from Arize-side failure.

After implementation:

Run the application and trigger at least one LLM call.

Use the arize-trace skill to confirm traces arrived. If empty, retry shortly. Verify spans have expected openinference.span.kind, input.value/output.value, and parent-child relationships.

If no traces: verify ARIZE_SPACE and ARIZE_API_KEY, ensure tracer is initialized before instrumentors and clients, check connectivity to otlp.arize.com:443, and inspect app/runtime exporter logs so you can tell whether spans are being emitted locally but rejected remotely. For debug set GRPC_VERBOSITY=debug or pass log_to_console=True to register(). Common gotchas: (a) missing project name resource attribute causes HTTP 500 rejections — service.name alone is not enough; Python: pass project_name to register(); TypeScript: set "model_id" or SEMRESATTRS_PROJECT_NAME on the resource; Go: add attribute.String("openinference.project.name", "my-app") to resource.New(...); (b) CLI/script processes exit before OTLP exports flush — call provider.force_flush() then provider.shutdown() (Python/TS) or tp.Shutdown(ctx) (Go) before exit; (c) CLI-visible spaces/projects can disagree with a collector-targeted space ID — report the mismatch instead of silently rewriting credentials.

If the app uses tools: confirm CHAIN and TOOL spans appear with input.value / output.value so tool calls and results are visible.

When verification is blocked by CLI or account issues, end with a concrete status:

app instrumentation status

latest local trace ID or run ID

whether exporter logs show local span emission

whether the failure is credential, space/project resolution, network, or collector rejection

Emitting session.id for multi-turn session tracking

For session-level evaluations in Arize (e.g. the {conversation} template variable in arize-evaluator), spans must carry attributes.session.id. This section covers how to emit it correctly.

Pattern (Python / OpenTelemetry):

from typing import Optional
import uuid

def chat(question: str, session_id: str, chat_history: Optional[list] = None) -> str:
    with tracer.start_as_current_span("chat") as span:
        span.set_attribute("openinference.span.kind", "CHAIN")
        span.set_attribute("input.value", question)
        span.set_attribute("session.id", session_id)
        answer = your_llm_call(question)  # replace with the actual LLM call
        span.set_attribute("output.value", answer)
        return answer

Generate session_id once per conversation and pass it in from the caller:

session_id = str(uuid.uuid4())  # generate once per conversation, not per turn

for user_message in conversation:
    reply = chat(user_message, session_id=session_id)

Rules:

Same session_id for every turn in one conversation. Each call produces its own span, but all spans share the same session_id so Arize groups them.

New session_id when a fresh conversation starts. Generate a new UUID at the start of each session, not at each turn.

Pass session_id in from the caller — the caller (request handler, notebook cell, app frontend) owns the session boundary.

Auto-instrumentation note: If using OpenInference auto-instrumentation (e.g. for LiteLLM or OpenAI), you do not control span creation directly. Wrap the LLM call in a manually-created CHAIN span and set session.id there — the auto-instrumented LLM spans will nest under it as children.

Flushing spans (Jupyter notebooks and short-lived scripts):

The OTLP batch exporter holds spans in memory and ships them on a timer or when the buffer fills. In a Jupyter notebook or short-lived script the process may end before the buffer flushes — spans emit correctly locally but never reach Arize.

Call force_flush() after finishing a conversation you want to inspect:

tracer_provider.force_flush()   # ships buffered spans immediately

Call shutdown() only when done tracing entirely — it closes the exporter and cannot be reversed without re-initializing:

tracer_provider.force_flush()
tracer_provider.shutdown()

Checklist before debugging missing session data in Arize:

Is session.id set on the CHAIN span? If using auto-instrumentation, wrap the LLM call in a manual CHAIN span and set it there.

Is the same session_id passed to every turn in the conversation?

Was force_flush() called after the conversation ended?

Did you wait ~30 seconds for ingestion before checking the Arize UI?

Reference links

Resource
URL

Agent-Assisted Tracing Setup
https://arize.com/docs/ax/alyx/tracing-assistant

Agent Setup Prompt (full routing + phases)
https://arize.com/docs/PROMPT.md

Arize AX Docs
https://arize.com/docs/ax

Full integration list
https://arize.com/docs/ax/integrations

Doc index (llms.txt)
https://arize.com/docs/llms.txt

IDE Integration (MCP)

If the user asks about IDE-based instrumentation guidance or MCP setup, see references/tracing-assistant-mcp.md.

Save Credentials for Future Use

See references/ax-profiles.md § Save Credentials for Future Use.