Agent harnesses

An open-source agent runtime shell from DeepSeek using a fully plugin-based architecture with a web interface, turning language models into executors for multi-step tasks for developers building assistants.

Adds a runtime enhancement shell around popular coding agents, enabling multi-model orchestration, persistent memory, and graph-based engineering workflows through simple keyword-triggered prompts.

Official plugin allowing Claude Code users to delegate tasks and request code reviews from Codex, with background task management support.

Adds a runtime shell around Codex agents with hooks, agent teams, HUDs and workflow enhancements, helping developers coordinate and manage coding assistant sessions.

Adds an enhancement shell around the Codex desktop app, streamlining everyday workflows, navigation, and operation for developers seeking a more comfortable and efficient coding-assistant experience.

A game development kit for Claude Code that provides forty-nine specialized expert agents and coordinated workflows to help developers structure AI-assisted game creation.

Optimizes coding-agent context windows by compressing tool output, persisting session memory, and routing requests across multiple model platforms through MCP and hooks.

Configuration framework enhancing Claude Code with specialized commands, cognitive personas, development methodologies, and workflow automation for structured software development teams.

A desktop GUI toolkit for Claude Code that lets developers manage sessions, create custom agents, run background tasks, and track usage statistics.

Life operating system running on agents like Claude Code, providing memory, skills, context, and intent-engineering enhanced shell for personal productivity management.

Lightweight web interface for the Hermes agent, offering three-column chat and workspace browsing that matches the command-line experience for remote users.

Runtime harness connecting LLMs to real browsers, enabling external agents to operate the web autonomously while evolving and accumulating reusable assistants and behaviors.

Lightweight agent runtime shell with a built-in personal assistant, offering dozens of tools, skill memory, and multi-agent collaboration for users.

Obsidian plugin for knowledge workers that embeds Claude Code and similar coding agents as in-vault collaborators, bringing AI assistance directly into note-taking and organization workflows.

A multiplayer agent harness for work that provides isolated workspaces, memory, sandboxes, and background tasks, supporting multiple models and carriers with collaborative runtime management.

Engineering shell for AI coding agents that persists requirements, tasks, and memory to support standardized development workflows across 22 platforms.

Local-first cross-platform desktop workspace enhances Claude Code and agent workflows with multi-agent sessions, Git worktrees, code diffs, skill marketplace, multi-model access, and messaging integrations.

Production-grade multi-agent runtime where a Queen coordinator directs cloned workers executing business processes in parallel, with state management, recovery, and supervision for enterprise operators.

Command-line tool for installing, syncing, and lazily loading portable SKILL.md files across different AI coding agents, enabling developers to share skills and optimize context usage.

Desktop and web workspace for the OpenCode AI agent, enabling developers to manage sessions and goals, run multi-model comparisons, and review changes before publishing.

A development scaffolding system for coding agents like Claude Code, providing context engineering patterns and staged delivery loops to structure incremental implementation.

Autonomous development loop wrapper for Claude Code that iteratively improves a project until completion, with smart exit detection and rate-limit protection for developers.

A GEP-based self-evolution engine for AI agents that consolidates scattered prompt-tuning experience into auditable, reusable gene and capsule assets for continuous improvement.

Provides a lean OpenCode multi-agent suite that mixes models and automatically delegates coding tasks to specialized subagents running in parallel.

Local-first agentic HTML editor that reuses an authenticated CLI agent and 75 skill templates to generate magazines, slides, posters and six other page types, with sandboxed preview and one-click export.

Provides a local browser shell for reviewing coding-agent plans and diffs visually, annotating feedback, sharing with teammates, and sending comments back with one click.

An AI validation proxy for git push that runs checks in an isolated workspace and automatically opens clean pull requests for developers.

Agent engineering kit for Antigravity providing rules, skills, memory, orchestration, and safety hooks inside a runnable shell for developers building controlled agent workflows.

TypeScript library for orchestrating sandboxed coding agents, providing isolated execution environments with branching and merge capabilities for safe parallel development workflows.

Desktop graphical interface for OpenClaw agents that enables chatting, channel management, skill configuration and scheduled tasks without using the terminal.

A runtime injector and reasoning-mode routing suite that enhances DSH agents with smarter task execution.

Provides a deterministic engineering environment for existing coding agents, offering shared memory, structured workflows, and evidence trails to improve collaboration reliability for development teams.

Agent scaffolding for autonomous game development with Godot and Bevy, turning requirements into repository templates that drive Claude Code and Codex to build and verify games.

Visual plugin for IntelliJ-based IDEs wrapping coding CLIs such as Claude Code and Codex, adding chat, context handling, session management, and agent-task controls.

Provides a multi-agent offensive-security framework that turns existing coding agents into autonomous red-team workers for reconnaissance, exploitation, and reporting to find zero-day vulnerabilities.

Zero-configuration scaffolding tool for Claude Code and Codex that automates initial setup by installing workflows, agent systems, and ready-to-use API plus MCP integrations for AI coding projects.

Runtime safety guard for users of AI coding agents, blocking dangerous git and shell commands from execution with simple installation and configurable protection rules.

Adaptive AI-driven development lifecycle workflow rules with a neutral core that runs across mainstream coding-assistant harnesses to steer agent behavior.

Implements a self-organizing Obsidian vault that gives coding agents persistent memory by managing notes, indexes, and continuing session context across tools.

Versioned deployment manager for Codex instructions offering preview, ownership manifests, hook isolation, scenario evaluation, and recovery for teams maintaining prompts.

A multi-engine desktop GUI that wraps CLI coding agents like Claude Code and Codex with chat, session history, and dev panels.

Cascading runtime layer embedded in the agent loop that optimizes cost, latency, quality, and policy decisions during model calls while integrating with multiple agent frameworks.

Context management shell for Claude Code that maintains development state through ledgers and handoffs, reduces context pollution during MCP execution, and orchestrates agents with isolated windows.

A local context-engineering layer for AI coding agents that handles context routing, prompt compression, session memory, and cost tracking to keep large codebases within token budgets.

Framework of rules, sub-agents, and skills that configures Claude Code as a general-purpose offensive and defensive security agent, orchestrating security tools for adversarial research and operations.

Extension for Pi enabling asynchronous subagent delegation with output truncation, artifact handling, and session sharing to support review and parallel agentic workflows.

Agent harness for complex codebases that adds project memory, structured planning, task execution, and verified completion on top of Codex workflows for developers maintaining large repositories.

An agentic orchestration harness that equips coding agents with a full dev team.

Agent skill harness for turning ideas into evaluated coding workflows, offering dual-mode runtime, memory, and distribution with assessment for long-running development tasks.

Operation-layer plugin suite for the Hermes coding agent providing problem modeling, workflow orchestration, and long-term memory management for developers building custom workflows.

Collective skill evolution system for multi-agent setups that captures experience from real conversations and redeploys reusable skills. Enables shared learning and automatic skill distribution.

Wraps the Pi agent framework as a lightweight coding harness bundled with extensions and skills, optimizing cold startup and usability for smaller local language models.

A browser extension that brings MCP tools to web-based AI assistants such as ChatGPT, Gemini, Grok, and DeepSeek. It detects available tools and enables direct execution inside chat pages.

Harness analyzes project and session evidence from coding agents to surface loop-level insights, prioritized improvements, and verifiable next steps for developers refining agent workflows.

A local-first desktop app for agent developers to prototype ideas, inspect each harness step, replay failures, and evaluate performance, with cloud support for managed agents.

A polished desktop shell for DeepSeek Harness with bundled runtime, themes, plugin marketplace, and safety guards.

Gateway that gives cloud AI coding assistants a real development environment on local machines, letting them read and edit code, run tests, and use Git.

Maintains a continuous background code-review database for AI coding agents, feeding issues back with accountability and automatic fixes to protect generated code quality.

Equips AI coding agents with a modular, stack-agnostic set of security-review skills for threat modeling, vulnerability reproduction, and automated patching across software supply chains.

Control-plane shell layer unifying multiple AI coding agents, providing centralized configuration, session management, local memory, and reusable engineering skills for software development workflows.

Provides a deployment and operations harness for OpenClaw with setup wizards, watchdog supervision, rollback support, and multi-channel management for long-running multi-agent installations.

Wraps Claude Code with a verification-driven development harness that manages planning, execution, and synchronization, enforcing quality gates plus model and effort routing from a single Go binary.

Telegram-based mobile client for OpenCode that runs coding sessions locally while enabling remote monitoring and control from phones. Supports model switching, task scheduling, and progress tracking for developers working away from their machines.

A red-team runtime enhancement for Codex-style agents that adds target contracts, workflow orchestration, evidence acceptance checks, and terminal adjudication for adversarial testing.

Multi-layer runtime enhancement framework for large-model coding agents, combining refusal suppression, file routing, bundled MCP safety tools, skill modules, and Kali backend integrations.

Always-on Slack-based system that maximizes Claude Code Pro utilization by automatically processing queued tasks. Manages isolated workspaces, creates Git commits and pull requests, and balances day-night usage thresholds for development teams.

A management tool for deploying Claude Code instructions with safe CLI recovery workflows, plus an early desktop beta for macOS and Windows aimed at team administrators.

Adds a macOS voice assistant layer to Claude Code, letting users brainstorm projects by voice, trigger builds and receive spoken alerts when a coding session requires human attention, without extra API keys.

Universal desktop interface for running Codex, Claude Code, and other local AI agents side by side. Provides shared chat sessions, file inspection, browser context, and manual takeover controls for developers supervising parallel coding workflows.

Workflow harness that layers skill routing and delivery verification over mainstream coding agents. It helps developers delegate tasks that continue autonomously through implementation and validation until completion.

A polished web interface for the Claude Code and Codex CLIs, managing multiple sessions, project groups, transcript rendering, and plan-mode approval workflows.

A desktop harness for DeepSeek agents that provides runtime shell, skills and plugin extensions for daily use.

Provides an open-source, locally runnable multi-agent runtime compatible with CMA, supporting multiple models, MCP tools, scenario templates, session sandboxes, audit memory, and a visual dashboard.

A local-first managed runtime for AI agents with persistent sessions, sandboxed execution, MCP governance, audit, and replay.

A research codebase implementing an organizational control layer for agents, with approval, rewriting, and interception of behaviors plus traceable benchmarks for control decisions.

Orchestration layer for Codex that assigns planning, review, design, and execution roles to different models and coordinates their collaboration through goals and reusable workflows.

A ready-to-run desktop bundle of DeepSeek Harness with the official runtime, plugin marketplace, files, git, and terminal in one app.

A research infrastructure shell for coding agents that standardizes literature retrieval, citation, drafting, and validation into a stable academic workflow.