PH

phoenix-evals

Tools for creating and executing evaluation suites for AI and LLM-based applications.

Install

mkdir -p .claude/skills/phoenix-evals && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/1054" && unzip -o skill.zip -d .claude/skills/phoenix-evals && rm skill.zip

Installs to .claude/skills/phoenix-evals

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Build and run evaluators for AI/LLM applications using Phoenix.
63 charsno explicit “when” trigger
Advanced

Key capabilities

  • Build and run evaluation pipelines for LLM outputs
  • Compare model performance across different prompts
  • Automate quality checks for AI responses
  • Validate evaluator accuracy against human labels
  • Integrate evaluations into CI/CD test runners

How it works

The skill provides a framework to build deterministic code-based evaluators and LLM-based judges, allowing users to gate CI/CD pipelines based on quality signals.

Inputs & outputs

You give it
LLM application traces or outputs
You get back
Performance metrics and quality benchmarks

When to use phoenix-evals

  • Run eval suites on LLM outputs
  • Compare model performance across different prompts
  • Automate quality checks for AI responses

About this skill

Phoenix Evals

Build evaluators for AI/LLM applications. Code first, LLM for nuance, validate against humans.

Quick Reference

TaskFiles
Setupsetup-python, setup-typescript
Decide what to evaluateevaluators-overview
Choose a judge modelfundamentals-model-selection
Use pre-built evaluatorsevaluators-pre-built
Build code evaluatorevaluators-code-python, evaluators-code-typescript
Build LLM evaluatorevaluators-llm-python, evaluators-llm-typescript, evaluators-custom-templates
Batch evaluate DataFrameevaluate-dataframe-python
Run experimentexperiments-running-python, experiments-running-typescript
Run evals in a test runner (CI gate)integrations-pytest, integrations-vitest-jest
Create datasetexperiments-datasets-python, experiments-datasets-typescript
Generate synthetic dataexperiments-synthetic-python, experiments-synthetic-typescript
Validate evaluator accuracyvalidation, validation-evaluators-python, validation-evaluators-typescript
Sample traces for reviewobserve-sampling-python, observe-sampling-typescript
Analyze errorserror-analysis, error-analysis-multi-turn, axial-coding
RAG evalsevaluators-rag
Avoid common mistakescommon-mistakes-python, fundamentals-anti-patterns
Productionproduction-overview, production-guardrails, production-continuous

Workflows

Starting Fresh: observe-tracing-setuperror-analysisaxial-codingevaluators-overview

Building Evaluator: fundamentalscommon-mistakes-python → evaluators-{code|llm}-{python|typescript} → validation-evaluators-{python|typescript}

RAG Systems: evaluators-rag → evaluators-code-* (retrieval) → evaluators-llm-* (faithfulness)

Gating CI: evaluators-{code|llm}-{python|typescript} → integrations-{pytest|vitest-jest} → production-continuous

Production: production-overviewproduction-guardrailsproduction-continuous

Reference Categories

PrefixDescription
fundamentals-*Types, scores, anti-patterns
observe-*Tracing, sampling
error-analysis-*Finding failures
axial-coding-*Categorizing failures
evaluators-*Code, LLM, RAG evaluators
experiments-*Datasets, running experiments
integrations-*Run evals from test runners (pytest, Vitest, Jest) as a CI gate
validation-*Validating evaluator accuracy against human labels
production-*CI/CD, monitoring

Key Principles

PrincipleAction
Error analysis firstCan't automate what you haven't observed
Custom > genericBuild from your failures
Code firstDeterministic before LLM
Validate judges>80% TPR/TNR
Binary > LikertPass/fail, not 1-5
Invariants gate, signals trendassert/expect hard invariants (CI red); log LLM-judge quality signals and gate the aggregate (acceptance criteria), not every case

When not to use it

  • Tasks not involving LLM output evaluation

Prerequisites

Phoenix serverPython or TypeScript environmentOpenAI or Phoenix client packages

Limitations

  • Requires validation of judge models to ensure accuracy
  • Binary pass/fail is preferred over Likert scales

How it compares

It moves beyond manual review by providing a code-first framework to automate quality signals and track performance trends.

Compared to similar skills

phoenix-evals side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
phoenix-evals (this skill)329dNo flagsAdvanced
agent-evaluation36moNo flagsAdvanced
mlops-validation26moNo flagsIntermediate
evaluating-machine-learning-models127dReviewIntermediate

Try saying

Example prompts that trigger this skill in your AI assistant.

Search skills

Search the agent skills registry