LLM 可观测性

工具排名

按实测增长排序，并在不同来源间归一化。 原始值保留各自的数据窗口；计算变化需要 7 天内至少两次测量。

1. 1

   agent-stack

   技能

   活跃最近提交：3 天前 · 2026-09-01

   Production patterns for agent orchestrators, harnesses, evals, MCP/A2A interoperability, memory, provider routing, and LLM usage metering.

   Python许可证: MIT

   • a2a
   • agent-orchestration
   • agent-skills
   • ai-agents
   • developer-tools

   github30 天前创建实测增长打开来源 ↗

   安装 git clone https://github.com/ssheleg/agent-stack ~/.claude/skills/agent-stack

   2GitHub 星标+1 (+100.0 %)
2. 2

   untell

   其他

   活跃最近提交：今天 · 2026-09-05

   AI-detector auditing toolkit: measures how often a detector flags genuine human writing, how stable its verdict is across seeds, and whether that verdict survives meaning-preserving edits. Detector-in-the-loop measurement harness. Claude…

   Python许可证: MIT

   • ai-detection
   • ai-detector
   • benchmark
   • claude-code
   • claude-code-skill

   github2 个月前创建实测增长打开来源 ↗

   18GitHub 星标稳定
3. 3

   opik-claude-code-plugin

   MCP

   活跃最近提交：37 天前 · 2026-07-30

   Log Claude Code sessions to Opik, the open-source LLM observability and evaluation platform, built by Comet. Tracing, evaluation, and skills for observable AI applications.

   Go许可证: Apache-2.0

   • ai-agent
   • claude-ai
   • claude-code-plugin
   • llm-observation
   • mcp-server

   github6 个月前创建实测增长打开来源 ↗

   22GitHub 星标+1 (+4.8 %)
4. 4

   bar-observatory

   MCP

   活跃最近提交：8 天前 · 2026-08-27

   Deterministic, local-only audit reports for Claude Code AI agent sessions. Rust + SQLite. Zero network calls. MCP server for agents.

   HTML许可证: MIT

   • agent-monitoring
   • agentic-ai
   • ai-agents
   • ai-engineering
   • audit

   github23 天前创建最新实测增长打开来源 ↗

   22GitHub 星标+2 (+10.0 %)
5. 5

   deslop-GPT

   技能

   活跃最近提交：1 天前 · 2026-09-04

   Deletion-first Agent Skill for removing test bloat, verification theater, and speculative fallbacks while preserving behavior.

   Python许可证: MIT

   • agent-skills
   • claude
   • claude-ai
   • claude-code
   • claude-skills

   github12 天前创建最新实测增长打开来源 ↗

   安装 git clone https://github.com/MrZoyo/deslop-GPT ~/.claude/skills/deslop-GPT

   87GitHub 星标+59 (+210.7 %)
6. 6

   agenttap

   智能体

   休眠最近提交：5 个月前 · 2026-04-05

   Real-time debugging proxy for Agent2Agent (A2A) multi-agent systems

   许可证: MIT

   • a2a
   • agent2agent
   • debugging
   • debugging-tool
   • developer-tools

   github5 个月前创建实测增长打开来源 ↗

   3GitHub 星标稳定
7. 7

   agentanvil

   智能体

   休眠最近提交：4 个月前 · 2026-04-29

   Contract-based testing for LLM agents: hybrid evaluation, multi-agent + A2A, deterministic replay.

   Python许可证: MIT

   • a2a
   • agents
   • contracts
   • evaluation
   • llm

   github4 个月前创建实测增长打开来源 ↗

   0GitHub 星标稳定
8. 8

   sentry-instrumentation

   技能

   活跃最近提交：2 个月前 · 2026-06-19

   Sentry instrumentation skill for system-behavior tracking

   Shell许可证: MIT

   • agent-skills
   • claude-skills
   • codex-skill
   • instrumentation
   • monitoring

   github4 个月前创建实测增长打开来源 ↗

   安装 git clone https://github.com/tortastudios/sentry-instrumentation ~/.claude/skills/sentry-instrumentation

   24GitHub 星标稳定
9. 9

   adaptive-harness

   技能

   休眠最近提交：5 个月前 · 2026-03-16

   A self-improving harness router for Claude Code.

   Python许可证: MIT

   • a-b-testing
   • ai-agent
   • ai-orchestration
   • ai-routing
   • ai-workflow

   github5 个月前创建实测增长打开来源 ↗

   安装 /plugin marketplace add SeongwoongCho/adaptive-harness

   8GitHub 星标稳定
10. 10

    axiom

    技能

    活跃最近提交：5 天前 · 2026-08-31

    Axiom is a curated marketplace of shared plugins for Claude Code and Codex.

    Python许可证: MIT

    • agent-skills
    • anthropic
    • claude
    • claude-code
    • claude-code-marketplace

    github3 个月前创建实测增长打开来源 ↗

    安装 git clone https://github.com/netopsengineer/axiom ~/.claude/skills/axiom

    5GitHub 星标稳定
11. 11

    astragraph

    智能体

    休眠最近提交：5 个月前 · 2026-03-16

    Policy-enforced observability and fail-closed guardrails for MCP/A2A multi-agent systems.

    Rust

    • a2a
    • ai-agents
    • llm-security
    • mcp
    • multi-agent

    github7 个月前创建实测增长打开来源 ↗

    26GitHub 星标稳定
12. 12

    CodeFlow

    智能体

    活跃最近提交：2 天前 · 2026-09-03

    面向长程研发任务的 Java Agent Harness，支持 A2A 跨语言协作、可恢复执行、上下文工程与 Eval 驱动开发。

    Java许可证: Apache-2.0

    • a2a
    • agentic-ai
    • ai-agents
    • developer-tools
    • evals

    github10 天前创建最新实测增长打开来源 ↗

    0GitHub 星标稳定
13. 13

    kaggle-capstone-ai-agent

    智能体

    休眠最近提交：9 个月前 · 2025-11-26

    A safety-first multi-agent mental health companion with real-time distress tracking, triple-layer guardrails, and evidence-based grounding techniques. Built for Kaggle × Google Agents Intensive 2025 Capstone (Agents for Good Track)

    Python

    • a2a-protocol
    • agent-evaluation
    • agents-for-good
    • ai-agents
    • crisis-support

    github9 个月前创建实测增长打开来源 ↗

    1GitHub 星标稳定
14. 14

    gt8004-sdk

    其他

    休眠最近提交：6 个月前 · 2026-02-23

    Official Python SDK for GT8004 — AI agent observability with MCP, A2A, x402 payment tracking. FastAPI, Flask, FastMCP middleware included.

    Python许可证: MIT

    • a2a
    • ai-agent
    • analytics
    • fastapi
    • flask

    github6 个月前创建实测增长打开来源 ↗

    1GitHub 星标稳定
15. 15

    langsmith-fetch-skill

    技能

    休眠最近提交：5 个月前 · 2026-04-06

    🔍 AI observability skill for Claude Code. Debug LangChain/LangGraph agents by fetching execution traces from LangSmith Studio directly in your terminal.

    许可证: MIT

    • agent-debugging
    • ai-agents
    • claude-skills
    • claude-skills-creator
    • claude-skills-hub

    github8 个月前创建实测增长打开来源 ↗

    安装 git clone https://github.com/OthmanAdi/langsmith-fetch-skill ~/.claude/skills/langsmith-fetch-skill

    27GitHub 星标稳定
16. 16

    mcp-smb-products

    MCP

    活跃最近提交：2 个月前 · 2026-07-03

    8 MCP SMB products — standalone AI servers for SMBs: Guardrails, FinOps, Observability, Router, Trust Score, Memory, ThinkSecure, A2A Lite. 37 tools, TNC credits billing.

    Python许可证: MIT

    • a2a-protocol
    • ai-governance
    • finops
    • guardrails
    • mcp

    github4 个月前创建实测增长打开来源 ↗

    1GitHub 星标稳定
17. 17

    agentops

    智能体

    休眠最近提交：4 个月前 · 2026-04-13

    AgentOps: Multi-agent infrastructure remediation platform. A2A protocol, agent coordination, HITL approval, auto-rollback.

    Python许可证: MIT

    • a2a-protocol
    • ai
    • automation
    • infrastructure
    • llm

    github6 个月前创建实测增长打开来源 ↗

    0GitHub 星标稳定
18. 18

    agent-inspector

    技能

    休眠最近提交：7 个月前 · 2026-01-15

    Local open-source dev tool to debug, secure, and evaluate LLM agents. Provides static analysis, dynamic security checks, and runtime monitoring - integrates with Cursor and Claude Code.

    Python

    • agent-debugging
    • agent-security
    • agent-trace
    • ai-security-tool
    • behavior-analysis

    github10 个月前创建实测增长打开来源 ↗

    安装 /plugin marketplace add cylestio/agent-inspector

    9GitHub 星标稳定
19. 19

    AgentStack

    智能体

    活跃最近提交：1 天前 · 2026-09-04

    Provide clear documentation for AgentStack’s MCP protocol, plugins, and ecosystem API with usage examples and tool references.

    • a2a
    • anthropic
    • claude-code
    • crewai
    • deep-learning

    github5 个月前创建实测增长打开来源 ↗

    0GitHub 星标稳定
20. 20

    Agent-Reliability-and-Evaluation-Lab

    MCP

    活跃最近提交：15 天前 · 2026-08-21

    Production reliability, benchmarking, and evaluation layer for MCP agent runtimes — protocol negotiation, A2A interop, circuit breakers, connection pooling, adaptive streaming, and autoscaling, with measured throughput/latency and a…

    Python许可证: Apache-2.0

    • a2a-protocol
    • agent-to-agent
    • ai-agents
    • async-python
    • autoscaling

    github18 天前创建最新实测增长打开来源 ↗

    3GitHub 星标稳定
21. 21

    opus-fable-playbook

    技能

    活跃最近提交：2 个月前 · 2026-07-03

    Make Claude Opus 4.8 behave like Claude Fable 5 — doctrine output style, drift-catching hooks, and an eval loop against golden Fable transcripts. Claude Code plugin.

    Shell许可证: MIT

    • anthropic
    • claude
    • claude-code
    • claude-code-plugin
    • evals

    github2 个月前创建实测增长打开来源 ↗

    安装 /plugin marketplace add rennf93/opus-fable-playbook

    34GitHub 星标+1 (+3.0 %)
22. 22

    adherencia-reglas

    技能

    活跃最近提交：36 天前 · 2026-07-31

    Mide si tu agente de IA cumple las reglas que le escribiste. Lee el historial local de Claude Code y devuelve un porcentaje por regla. Sin instalar nada, sin red, biblioteca estándar.

    Python许可证: MIT

    • agent-evaluation
    • ai-agents
    • claude-code
    • claude-code-skill
    • claude-code-skills

    github40 天前创建实测增长打开来源 ↗

    安装 git clone https://github.com/jleonceo/adherencia-reglas ~/.claude/skills/adherencia-reglas

    1GitHub 星标稳定
23. 23

    claudestat

    技能

    活跃最近提交：3 天前 · 2026-09-02

    Real-time execution trace and cost intelligence for Claude Code

    TypeScript许可证: MIT

    • agentic-workflow
    • ai-productivity
    • anthropic
    • anthropic-claude
    • claude

    github4 个月前创建实测增长打开来源 ↗

    安装 /plugin marketplace add DeibyGS/claudestat

    34GitHub 星标稳定
24. 24

    skill-receipts

    技能

    活跃最近提交：37 天前 · 2026-07-30

    Claude Code skills where every entry ships receipts — accuracy-gated benchmarks against baseline and placebo, rejects published

    JavaScript许可证: MIT

    • agent-skills
    • ai-agents
    • anthropic
    • benchmark
    • claude

    github45 天前创建实测增长打开来源 ↗

    安装 git clone https://github.com/sjh9714/skill-receipts ~/.claude/skills/skill-receipts

    2GitHub 星标稳定
25. 25

    green-agent

    智能体

    活跃最近提交：2 个月前 · 2026-07-01

    A2A green-agent orchestrator for evaluating agents on the AppWorld benchmark, built on the AgentBeats SDK

    Python

    • a2a-protocol
    • agentbeats
    • appworld
    • evaluation
    • llm-agents

    github11 个月前创建实测增长打开来源 ↗

    0GitHub 星标稳定