Eval
Evaluate and rank agent results by metric or LLM judge for an AgentHub session.
Before & after
Same prompt, same agent — the only difference is the skill.
Good question. I can help you evaluate the results. To compare them fairly, I'll need to see the outputs from all three versions and know what success looks like for your use case. Are you measuring accuracy, latency, token efficiency, user satisfaction, or something else? If you share the results and your success criteria, I can offer an analysis.
AGENT COMPARISON REPORT VERSION PERFORMANCE TABLE Metric | v1 Baseline | v2 Variant-A | v3 Variant-B | Winner Accuracy | 82.1% | 87.4% | 84.9% | v2 (+5.3pp) Latency (ms) | 1240 | 1520 | 890 | v3 (-350ms) Token Cost/Call | $0.042 | 0.051 | 0.038 | v3 (-9%) User Rating | 3.2/5 | 3.8/5 | 3.5/5 | v2 (+0.6) STATISTICAL SIGNIFICANCE Accuracy lift (v2 vs v1): +5.3pp, 95% CI: 3.1-7.5pp, p=0.002 Effect size: 0.47 (small-to-medium) Sample size: n=412 calls per version RISK ASSESSMENT v2 strength: Best accuracy and user preference v2 weakness: 22% slower (latency tradeoff) v3 strength: Fastest and cheapest per call v3 weakness: Accuracy gap vs v2 (-2.5pp) HEAD-TO-HEAD: v2 vs v3 To break the tie, run 200 calls at peak usage times Metric to watch: Does v3 latency advantage outweigh accuracy cost for production users? RECOMMENDATION Short term: Deploy v2 to 20% of traffic (explore accuracy gains) Long term: Hybrid approach – route high-stakes queries to v2, standard queries to v3 Expected outcome: +4% accuracy, -12% cost vs current baseline NEXT STEPS 1. Load test v2 at production scale (p99 latency target: 1500ms) 2. Collect user feedback on v2 output quality in staging 3. Define rollback criteria if accuracy variance exceeds 1.5pp
About this skill
name: eval description: Use when Evaluate and rank agent results by metric or LLM judge for an AgentHub session.
Eval
Evaluate and rank agent results by metric or LLM judge for an AgentHub session. Use when the user runs /hub:eval or asks to score, compare, or pick a winner among completed AgentHub agents.
What you get
- Public GitHub repo (alirezarezvani/claude-skills)
- the eval skill folder with SKILL.md. Part of a 337-skill / 30-agent / 70-command install.
Customize your output
- Fork the repo and adapt the skill's instructions and references to your workflow.
Example output
Activates automatically when your request matches Eval; chains with the other skills, agents, and commands in the collection.
Best for
Creators, builders, and teams using Claude Code.
SKILL.md preview
---
name: eval
description: Use this skill when the user runs the hub evaluation command or asks to score, compare, or pick a winner among multiple completed agent results from the same session.
version: 1.0.0
category: Development / Engineering
author: AgentVolt
license: proprietary
tags:
- development
- engineering
---
# Eval
Evaluates and ranks a set of completed agent results by a chosen metric or LLM-judge comparison, and returns a winner with the reasoning behind it.
## When to use
… (sign up to view the full skill)More development skills
View all Development skills →Generate
Scaffold code, tests, and boilerplate from a short spec.
Performance Profiler
Systematic performance profiling for Node.js, Python, and Go applications.
Threat Detection
Use when hunting for threats in an environment, analyzing IOCs, or detecting behavioral anomalies in telemetry.
Agent Harness
Turns a domain folder of skills into a bounded agentic loop: compile a goal into a task plan, execute with the domain's tools, verify every task, retry within a budget.