Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Build and run evaluators for AI/LLM applications using Phoenix.
.claude/skills/phoenix-evals/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 85% | 20 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-18 | ✓→✓ | ＝ Same ✓ | 25% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 34% | 0% |
Build evaluators for AI/LLM applications. Code first, LLM for nuance, validate against humans.
| Task | Files | | ---- | ----- | | Setup | setup-python, setup-typescript | | Decide what to evaluate | evaluators-overview | | Choose a judge model | fundamentals-model-selection | | Use pre-built evaluators | evaluators-pre-built | | Build code evaluator | evaluators-code-python, evaluators-code-typescript | | Build LLM evaluator | evaluators-llm-python, evaluators-llm-typescript, evaluators-custom-templates | | Batch evaluate DataFrame | evaluate-dataframe-python | | Understand experiments | experiments-overview | | Run experiment | experiments-running-python, experiments-running-typescript | | Create dataset | experiments-datasets-python, experiments-datasets-typescript | | Generate synthetic data | experiments-synthetic-python, experiments-synthetic-typescript | | Validate evaluator accuracy | validation, validation-evaluators-python, validation-evaluators-typescript | | Sample traces for review | observe-sampling-python, observe-sampling-typescript | | Analyze errors | error-analysis, error-analysis-multi-turn, axial-coding | | RAG evals | evaluators-rag | | Avoid common mistakes | common-mistakes-python, fundamentals-anti-patterns | | Production | production-overview, production-guardrails, production-continuous |
Starting Fresh: observe-tracing-setup → error-analysis → axial-coding → evaluators-overview
Building Evaluator: fundamentals → common-mistakes-python → evaluators-{code|llm}-{python|typescript} → validation-evaluators-{python|typescript}
RAG Systems: evaluators-rag → evaluators-code- (retrieval) → evaluators-llm- (faithfulness)
Production: production-overview → production-guardrails → production-continuous
| Prefix | Description | | ------ | ----------- | | 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 | | validation-* | Validating evaluator accuracy against human labels | | production-* | CI/CD, monitoring |
| Principle | Action | | --------- | ------ | | Error analysis first | Can't automate what you haven't observed | | Custom > generic | Build from your failures | | Code first | Deterministic before LLM | | Validate judges | >80% TPR/TNR | | Binary > Likert | Pass/fail, not 1-5 |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | pass→pass | 11,925 | 8,579 | -28% | 1 | 1 | 0% | 2,209 | 2,755 | +25% | 0 | 0 | — |
case-01 | fail→pass | 12,363 | 11,865 | -4% | 1 | 1 | 0% | 3,009 | 3,679 | +22% | 0 | 0 | — |
case-02 | fail→pass | 12,615 | 11,553 | -8% | 1 | 1 | 0% | 2,967 | 3,771 | +27% | 0 | 0 | — |
case-03 | fail→fail | 14,113 | 12,578 | -11% | 1 | 1 | 0% | 3,028 | 3,781 | +25% | 0 | 0 | — |
case-04 | pass→pass | 13,015 | 8,676 | -33% | 1 | 1 | 0% | 2,118 | 2,848 | +34% | 0 | 0 | — |
case-05 | pass→pass | 7,625 | 5,487 | -28% | 1 | 1 | 0% | 1,481 | 2,129 | +44% | 0 | 0 | — |
case-06 | fail→pass | 12,372 | 10,606 | -14% | 1 | 1 | 0% | 2,460 | 3,076 | +25% | 0 | 0 | — |
case-07 | pass→pass | 12,212 | 12,784 | +5% | 1 | 1 | 0% | 2,351 | 3,674 | +56% | 0 | 0 | — |
case-08 | pass→pass | 16,695 | 13,403 | -20% | 1 | 1 | 0% | 3,371 | 3,609 | +7% | 0 | 0 | — |
case-09 | pass→pass | 9,841 | 3,829 | -61% | 1 | 1 | 0% | 2,059 | 1,962 | -5% | 0 | 0 | — |
case-10 | pass→pass | 13,276 | 9,566 | -28% | 1 | 1 | 0% | 2,367 | 3,244 | +37% | 0 | 0 | — |
case-11 | pass→pass | 11,348 | 6,571 | -42% | 1 | 1 | 0% | 2,158 | 2,463 | +14% | 0 | 0 | — |
case-12 | pass→pass | 12,395 | 9,997 | -19% | 1 | 1 | 0% | 2,694 | 3,178 | +18% | 0 | 0 | — |
case-13 | pass→pass | 11,931 | 5,840 | -51% | 1 | 1 | 0% | 2,135 | 2,177 | +2% | 0 | 0 | — |
case-14 | pass→pass | 10,899 | 10,205 | -6% | 1 | 1 | 0% | 1,923 | 2,968 | +54% | 0 | 0 | — |
case-15 | pass→pass | 14,518 | 10,174 | -30% | 1 | 1 | 0% | 2,900 | 3,254 | +12% | 0 | 0 | — |
case-16 | pass→pass | 13,444 | 9,915 | -26% | 1 | 1 | 0% | 2,681 | 3,029 | +13% | 0 | 0 | — |
case-17 | pass→pass | 13,222 | 10,302 | -22% | 1 | 1 | 0% | 2,597 | 3,303 | +27% | 0 | 0 | — |
case-19 | pass→pass | 13,047 | 9,697 | -26% | 1 | 1 | 0% | 2,495 | 2,968 | +19% | 0 | 0 | — |
case-20 | pass→pass | 13,673 | 16,265 | +19% | 1 | 1 | 0% | 3,258 | 4,766 | +46% | 0 | 0 | — |
case-21 | pass→pass | 4,091 | 4,263 | +4% | 1 | 1 | 0% | 1,006 | 2,168 | +116% | 0 | 0 | — |
case-22 | pass→pass | 16,897 | 9,868 | -42% | 1 | 1 | 0% | 3,798 | 3,453 | -9% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +14 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 7/24/2026 | +27% |
Other measured skills in the registry, with their headline benchmark lift.