#human-in-the-loop

Review every AI edit before it touches your LaTeX. A local web app: comment on a passage, get a word-level diff, accept or reject. Uses Claude Code, Codex, Ollama or any OpenAI-compatible API. 本地运行的 LaTeX 论文 AI 审阅工具：批注 → diff → 接受。

VERDICT — a DFIR agent (Claude Code as the engine) that produces a signed, offline-verifiable verdict. SANS Find Evil! 2026.

Turn one Claude Code subscription into a crew of AI agents that research, plan, and ship code in parallel — a Slack-style HQ on your laptop where every risky action needs your approval. The HQ where AI agents work as a team, and you have the final say.

本地多运行时 Agent 工作台：围绕材料研究、生成可编辑报告、独立复核、追溯证据，并从人工反馈持续学习。

Local-first AI agent platform with a Tauri desktop app and Rust CLI. Run governed tools, durable sessions, multi-agent workflows, modular packages, and one-shot effect permits.

Reusable Interpretable Context Methodology (ICM) workspace template for filesystem-based AI agent workflows

Assign real work, share context, review output, and keep every human-agent handoff visible.

Human-in-the-loop planning plugin for Claude and Codex: turn ideas into reviewed Markdown plans, record decisions, enforce explicit execution gates, and provide tested controller primitives for safer agent workflows.

Local-first mission control for your AI agent workforce — run CLI agents like employees (SquadCue)

Open-source enterprise AI agent platform for centrally governing agents, tools, memory, RAG, approvals, and durable workflows with FastAPI, LangGraph, LangChain, and LangSmith.

Proxilion is the security layer for the agentic workforce. It turns managed AI agents into governed users by enforcing strict cryptographic boundaries on every API call to SaaS like Google Workspace, Salesforce, or Atlassian.

Policy-mapped evaluation lab for AI assistant behavior: approval gates, refusal boundaries, uncertainty handling, tool-use grounding, traces, and quality gates.

Playbook for PMs shipping AI products with PRDs, evals, HITL, launch gates, cost, and observability.

Agent + Multimodal RAG 全栈学习项目：LangGraph PAE/ReAct、Multi-Query、BM25 + pgvector 混合检索、RAG-Fusion/RRF、MinerU/OCR/ASR、evidence 引用与 HITL；React + Spring Boot + FastAPI。

Multi-Agent AI research platform using LangGraph for automated AI research. Orchestrate specialized agents (Planner, Researcher, Writer, Human-in-the-loop). Features a FastAPI backend, parallel web/academic retrieval (Tavily/Arxiv), and a React frontend for generating cited, grounded AI answer.

CLI and Preact kit that renders AI agent plans and reports as local HTML, with in-browser feedback sent back to the agent.

Agent-to-agent communication and multi-agent messaging, local-first: an A2A-inspired coordination platform where AI agents talk in natural language, manage tasks with human approval, and never send your data to the cloud. 65-command API, Python client, web UI, encrypted backups.

Agent personnel autonome en Rust, tournant sans écran sur un Mac. Telegram ou SSH, mémoire en Markdown, serveurs MCP, workflows durables, approbation avant tout ce qui engage.

Architectural taste for TypeScript coding agents. Minimize conceptual surface area, create boundaries deliberately, and generate code humans can navigate, review, and evolve.

Agent-safe action and workflow runner for TypeScript backends.

Skein is an agentic AI platform that answers questions by weaving together structured and unstructured data, grounding every response in a knowledge graph, hybrid vector search, and a governed semantic layer. It escalates to humans when unsure, and runs free locally or scales to the cloud.

Production-shaped, self-evolving AI agent operations platform: immutable agents, bounded loops and graphs, HarnessDev evolution, evaluations, observability, approvals, guardrails, and Cloudflare D1 deployment.

LLM-discovered legacy banking workflows compiled into typed capabilities for deterministic, policy-gated replay with same-session human intervention.

Governed actions, human approval, durable execution, reconciliation and audit for AI agents acting on real business systems.

Spec-driven development workflow for Claude Code — built around an approved design doc, where each stage waits for your approval.

Human-in-the-loop review for agent-written markdown. Self-hostable, Apache-2.0.

AI-powered CRM copilot with knowledge-grounded replies, sales hints and anti-hallucination safeguards.

Claude Code skill for evidence-backed social media planning, approval-first scheduling, publishing workflows, and analytics.

AI-powered Statement of Work drafting tool with a Human-in-the-Loop review gate, built with React and the Anthropic Claude API.

Agentic Video System — 17-agent LangGraph orchestrator for video editing: 12 laws, human-in-the-loop CEO approvals, LLM brain (OpenAI-compatible/Zhipu GLM), RAG knowledge repository, FastAPI + mermaid web dashboard, CLI. Plus Agenticine (OpenCode) agent config.

A decision library for AI agent sessions. Agents submit plans and ADRs over MCP; you read, comment, and green- or red-light them; every verdict — including every rejection — becomes memory the agents can query. Local-first, MIT.

Multi-agent LinkedIn branding system with ReAct tool-calling, human-in-the-loop approval, and full agent traces

AI-assisted collections & back-office automation console: risk scoring, human-approved workflows, and audit trail for overdue accounts. Next.js 15 + TypeScript.

Reusable agent skills for evidence-backed research, approval-gated publishing, schema-validated records, and durable context handoffs.

Your agent ships. You stay dangerous. A demanding code coach for Claude Code and Codex.

A privacy-aware multi-agent career management system with structured source governance, human approval workflows, and LLM output evaluation.

Technical case study of a production AI-powered lead qualification system built with WhatsApp, Claude, Whisper, Supabase, and Cloudflare Workers.

Human-gated AI UX pipeline that takes a BRD through research, requirements challenge, ideation, design, prototyping and synthetic validation using specialized agents and reusable skills.

A LangGraph agent backend built around production patterns: a streaming HTTP API, durable state that survives restarts, a human-approval gate for side-effecting tools, deterministic tests that never call a real LLM, and built-in observability.

中文版｜用 6 个固定故障场景，比较 Agent 在超时、限流、审批和重启后的行为。