agent-evals 开源项目推荐（116 个）

darkrishabh /agent-skills-eval

agentskills. io风格的人工智能代理技能的测试员

thetahealth /mirobody-eval

An evaluation harness for medical/health AI agents — reproduce and cover multiple benchmarks under one scoring discipline

thetahealth /mirobody-env

HAEnv — synthetic longitudinal patients with code-derived gold, for evaluating health agents over time

AgentX-ai /AgentX-Trace-Eval

The trust layer for your AI agent on any platform. Trace and evaluate your agent. Full Observability. Run Model portability analysis. Autotune your agent.

elm-os /rightmodeler

Replay real agent traces through cheaper models to prove which swaps are safe.

HumphreySun98 /repoagentbench

SWE-bench for your codebase — mine your merged PRs into local, contamination-free coding-agent benchmarks. Adapters: claude-code, aider (Opus 4.7 / GPT-5.5 / Sonnet 4.6 / Gemini 3.1 Pro).

MrTsepa /autoevolve

AI agent evolving strategies through automated self-play overnight. Generic framework with GEPA-inspired feedback loop and Elo tracking.

LMikeH /whileai-sdk

Scientific RL and SFT post-training for AI agents: build evals that can fail, simulate situations, judge checked against people, pass@1 with a 95% interval, train with GRPO/SFT/DPO, prove every gain on a held-out set. uv add whileai

The-Swarm-Corporation /StatisticalModelEvaluator

An implementation of the Anthropic's paper and essay on "A statistical approach to model evaluations"

tenurehq /GroundEval

Observe an agent run and GroundEval drafts the policy and diagram for you, no hand-written policy required, then scores what it checked, what it skipped, and what it wasn't allowed to touch.

iMeanAI /open-source-operator

创建你自己的开源运营者模型。

bitwise-media-group /evolve

A cli tool for evaluating coding agent plugins with a multi-tier approach.

hsaghir /looplet

Test-driven harness engineering for Python agents: own the loop, capture failures, and gate every change. Looplet is a domain-neutral, zero-dependency Python agent SDK, and harness engine.

agent-axiom /agent-anvil-leaderboard

Public Agent Anvil leaderboard submissions and generated index

Phoenix0531-sudo /BondLens

BondLens: evidence-first Chinese bond analysis agent — deterministic tools, optional LLM narration under guardrails, Trust Layer, live/snapshot/static data, Docker/CI.

shubchat /loab

LOAB: A benchmark for evaluating LLM agents on end-to-end mortgage lending operations under real regulatory constraints.

zwright8 /flight-recorder

Evidence, governance, and static reports for agentic runs

mverab /Reposcale

Alpha benchmark for repo continuation intelligence

salian /openknowledgebank

Portable, inspectable knowledge bundles for LLM agents.

bigkan8 /legal-action-boundary-eval

Legal Action Boundary Eval (LABE): public proxy eval for legal AI workflows at the action boundary