Run a reversible terminal coding agent with local or cloud LLMs.
Project overview
OpenHarness connects to local models such as Ollama or to cloud APIs, then reads, edits, and searches code, runs commands, uses MCP servers, delegates to sub-agents, and operates headlessly in CI. Permission modes, command-risk analysis, file checkpoints, automatic Git commits, and post-edit verification make agent changes easier to inspect and undo. It requires Node.js 18+.
Enforce AI-agent process and data-flow policies at the Linux kernel with eBPF.
Project overview
ActPlane applies labeled information-flow and causal-ordering rules below the tool layer, following an agent’s entire process tree, file edges, and network connections. It can block, kill, or notify on violations and feed a reason back to the agent; it requires Linux 5.8+ with BTF and elevated capabilities for the enforcement engine.
Drive coding CLIs through tmux with a normalized blocking interface.
Project overview
tmuxlet runs coding CLIs inside tmux and gives scripts one blocking interface that returns text or JSON while preserving an interactive terminal. It normalizes Claude, Codex, Gemini, opencode, pi, Cursor, and Grok, uses an explicit completion signal, and reports completed, blocked, timeout, or exited runs with inspectable logs. A default 24-session cap and optional memory guard limit fan-out. Rust, tmux, and at least one target CLI are required.
Schedule and chain autonomous agent skills with dashboards and notifications.
Project overview
AEON turns repository-based skill prompts into autonomous runs that can be scheduled, chained, and monitored from a dashboard. Results can be delivered through Telegram, Discord, Slack, or email, while each skill declares the API keys or MCP servers it needs. Setup requires Node.js 20+, an authenticated GitHub CLI, and your own repository copy; some skills depend on additional credentials.
Marks Crystal as deprecated and directs users to the Nimbalyst collaborative workspace.
Project overview
This repository states that Crystal was deprecated in February 2026 and directs users to Nimbalyst for active downloads, documentation, and updates. Nimbalyst is a local-first workspace for human and agent collaboration, with editors for code, Markdown, spreadsheets, diagrams, and visual documents. It uses Git worktrees to isolate parallel coding sessions and is built on Electron.
General skillCommunication and collaborationModel Context Protocol
riff
@NasonZ·Python
Let Claude and Codex collaborate as peers through delegated, consultative, or roundtable work.
Project overview
riff connects Claude Code with OpenAI’s Codex CLI as a peer collaborator rather than a subordinate tool. Its delegate, consult, and roundtable modes can use Codex MCP or `codex exec`, while a trace contract records context, constraints, execution, and acceptance criteria for non-trivial calls. Codex CLI is required; the MCP server is recommended for cleaner multi-turn exchanges.
Run an extensible coding agent on local models with a Python-only runtime.
Project overview
Claw Code Agent reimplements a Claude Code-style coding workflow in Python for OpenAI-compatible model servers. It provides a CLI, multi-turn chat, a local web interface, file and shell tools, sessions, permissions, plugins, and nested delegation while using only the Python standard library. Development is ongoing, and full MCP, hook, background-team, and platform parity is not yet complete.
Version an agent’s identity, rules, memory, tools, and skills in Git.
Project overview
Gitagent represents an AI agent as a Git repository, with model configuration, identity, rules, memory, tools, skills, hooks, and workflows stored as reviewable files. You can fork an agent, branch its personality, diff rule changes, and inspect memory history with normal Git operations. The runtime supports OpenAI, Anthropic, and OpenAI-compatible providers, and requires Node.js, Git, and the relevant API credentials.
Runs fresh-context agent loops from specs with commits and quality gates.
Project overview
py_ralph_frame provides a Python harness for spec-driven agent work: a plan defines the goal, specs define what to build, and each fresh-context iteration updates the implementation and project status. Pre-commit and pre-push gates can run linting, type checks, security scans, and tests for Claude, Codex, Gemini CLI, or Copilot; the harness intentionally does not create worktrees.
Develop in parallel across isolated Git worktrees with AI
Project overview
VibeTree gives each AI-assisted task its own Git worktree so several features or fixes can progress in parallel with separate terminal state. Desktop, browser, and mobile access are available, but the main branch is still developing cloud and broader platform capabilities; the README recommends release-v0.1 for a stable desktop-only build.
Move software ideas through specification, validation, security checks, and delivery.
Project overview
Project Maker guides Claude Code through staged, spec-driven development instead of one drifting conversation. It creates progressive on-disk artifacts for discovery, requirements, planning, execution, acceptance testing, security review, and pull-request delivery, while adapting the depth to the project size. Independent validators, resumable UAT, and explicit gates stop failed or risky work from being presented as complete.
Runs parallel Claude Code workspaces in containers with remote terminals, worktrees, and live previews.
Project overview
Catnip creates containerized workspaces with managed Git worktrees so multiple Claude Code sessions can run in parallel. Its web and native mobile interfaces, CLI, and SSH access expose full terminals, while port detection and a built-in proxy let you preview services from the workspace. It can run in Codespaces, devcontainers, or locally, but the current workflow is optimized for Claude Code and broader agent support is still planned.
Add tool governance, cost tracking, and audits to AI agents
Project overview
AutoHarness is a lightweight governance layer for existing AI agents. It can wrap an LLM client, check tool calls for risk and permissions, sanitize outputs, record decisions, and manage context and token budgets. YAML rules select different governance levels, while multi-agent profiles, per-call cost attribution, and session persistence cover longer-running workflows.
Run coding agents in parallel across isolated tmux workspaces
Project overview
amux is a terminal UI for running multiple coding agents in parallel. Each agent gets its own tmux session, and workspaces can use repository checkouts or Git worktrees so branch changes stay isolated. The same interface can launch agents, show diffs, and open a terminal; it supports Linux and macOS and requires tmux 3.2 or newer.
Manage coding-agent fleets across many repositories and worktrees from a Rust TUI.
Project overview
Run Claude Code, Codex, and Aider sessions across many repositories and worktrees from a durable tmux-backed terminal UI. Fleet, split, babysit, and focus views keep waiting agents visible, while timelines, session search, and an optional private-network bridge extend supervision. It needs tmux and Git; native Windows support is not available.
Turn recurring coding-agent mistakes into repository rules and checks.
Project overview
Harness Starter Kit helps an agent inspect a repository and add the smallest useful set of instructions, constraints, feedback loops, memory, checks, and evaluation records. Its prompt-first commands cover doctor, adopt, review, update, and refresh, while preserving existing architecture. The kit documents operating context; it does not by itself prove better agent outcomes.
Switch between multiple AI coding agents in one local CLI
Project overview
Nori is a native local CLI that puts Claude, Codex, Gemini, and other ACP-compatible agents behind one interactive terminal UI. It can reuse existing authenticated sessions on the machine, and the /agent command switches providers, making it useful for comparing tools without changing the surrounding workflow.
Explore a restricted experimental Rust coding-agent project
Project overview
Loom is an experimental Rust coding-agent research project with a REPL, file and code tools, server-side model proxying, conversation persistence, and remote execution components. Its repository explicitly restricts use to Geoffrey Huntley and warns that the software is unstable, unsupported, and subject to breaking API changes.