[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-76257":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":15,"stars30d":15,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":16,"rankGlobal":10,"rankLanguage":10,"license":17,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":20,"hasPages":18,"topics":21,"createdAt":10,"pushedAt":10,"updatedAt":31,"readmeContent":32,"aiSummary":33,"trendingCount":15,"starSnapshotCount":15,"syncStatus":34,"lastSyncTime":35,"discoverSource":36},76257,"reevesagents","mertkayacs\u002Freevesagents","mertkayacs","ReevesAgents is a Free and Open Source workspace manager for AI CLI agents. No API key needed. Tui WebUi Mcp Cli all in one package.","",null,"TypeScript",81,10,1,0,43.12,"Other",false,"master",true,[22,23,24,25,26,27,28,29,30],"ai-agents","autonomous-agents","claude","claude-c","gemini","llm-orchestration","multi-agent","openai-codex","tmux","2026-06-12 04:01:21","# reevesagents\n\nLocal tmux-first workspace manager for AI CLI terminals and agent teams.\n\nReevesAgents has two modes:\n\n- **Spawner mode:** the default. Start one tmux workspace with multiple independent provider CLI terminals. The human coordinates them. No MCP setup, no ReevesAgents environment injection, no root\u002Fworker roles.\n- **Orchestrator mode (BETA):** connected root\u002Fworker agents with MCP tools, messages, approvals, and agent spawning. This mode can write provider MCP config entries when you explicitly run setup.\n\nReevesAgents tracks each run in local state and opens every terminal or agent in its own tmux window. The registry is the source of truth; tmux is the execution and viewing surface. The TUI stays in the current or fallback `reeves` session, while each run gets its own tmux session to keep tabs uncluttered.\n\n## What It Does\n\n- Shows active, ended, and stale runs in a visible-menu TUI.\n- Starts multiple independent provider CLI terminals in Spawner mode.\n- Starts root and worker agents in Orchestrator mode (BETA).\n- Opens the real provider CLI window for each terminal or agent.\n- Lets Orchestrator agents and humans coordinate through MCP tools, CLI calls, and the TUI using run and agent ids.\n- Stores local JSON state under `~\u002F.reeves\u002Fruns`.\n- Registers ReevesAgents as an MCP server only when Orchestrator setup is explicitly run.\n- Uses a small animated Welcome screen, page-specific TUI layouts, and a static Runs dashboard to reduce terminal flicker.\n\nReevesAgents does not store provider credentials, proxy model traffic, embed a terminal emulator, or replace provider authentication.\n\n## Requirements\n\nCore runtime:\n\n- macOS, Linux, or WSL\n- Node.js 20.19+\n- tmux, 3.0+ recommended\n- A normal interactive shell on `PATH`\n\nProvider runtime:\n\n- At least one provider CLI for the provider you launch: `claude`, `codex`, `opencode`, or `hermes`\n- Provider authentication handled by that provider CLI\n\nMode-specific requirements:\n\n- Spawner mode needs only the core runtime plus the provider CLIs you launch.\n- Orchestrator BETA needs the same runtime, plus explicit MCP registration with `reevesagents orchestrator setup` if you want provider CLIs or host agents to call ReevesAgents tools.\n\nNot required:\n\n- Provider API keys in ReevesAgents config\n- Database, Docker, browser runtime, background service, or daemon\n- MCP setup for Spawner mode\n\nInstall is passive: no postinstall, no provider config writes, no background service.\n\n## Providers\n\n| Key | CLI | Launch | Notes |\n| --- | --- | --- | --- |\n| `cc` | Claude Code | `claude` | Supports model, API-key auth mode, effort, skip permissions, and Orchestrator BETA `\u002Fremote-control` startup injection |\n| `codex` | Codex CLI | `codex` | Supports model and skip permissions; Codex app-server remote control is managed outside agent launches |\n| `opencode` | OpenCode CLI | `opencode` | Supports `--prompt` and `--model`; ReevesAgents does not add undocumented skip flags |\n| `hermes` | Hermes | `hermes chat` | Supports model and `--yolo` skip permissions |\n\nDefault permissions are `ask`. Use `skip` only in trusted disposable workspaces.\n\n## Install From Source\n\n```sh\ngit clone https:\u002F\u002Fgithub.com\u002Fmertkayacs\u002Freevesagents.git\ncd reevesagents\npnpm install\npnpm build\npnpm link --global\n```\n\nVerify:\n\n```sh\nreevesagents --version\nreevesagents doctor\n```\n\n## Quick Start\n\n```sh\nreevesagents doctor\nreevesagents\n```\n\nThe first screen is always Welcome. It is a persistent main menu: use arrows and Enter to choose New Run, Runs, Doctor, Settings, Approvals, Reference, Credits, or Quit. New Run first asks for Spawner or Orchestrator BETA. When launched with run context, Welcome also shows Current Run. Runs auto-cleans ended and stale entries on refresh.\n\nSpawner mode can also start from the CLI:\n\n```sh\nreevesagents spawn codex:builder cc:reviewer --name \"release check\" --prompt \"Inspect the release state.\"\n```\n\nOrchestrator setup is explicit and BETA:\n\n```sh\nreevesagents orchestrator setup\n```\n\n## TUI Pages\n\n- Welcome: animated `REEVES AGENTS` block logo, chafa-rendered blocky duck mascot, and persistent main menu.\n- Runs: list all running, ended, and stale runs. Ended and stale runs auto-clean on refresh.\n- Run hub: show terminals for Spawner runs or root\u002Fworkers for Orchestrator BETA runs.\n- Terminal\u002Fagent detail: inspect provider, status, working directory, tmux ids, recent output, prompt\u002Ftask, open window, and close window.\n- New Run wizard: choose Spawner or Orchestrator BETA, then configure providers, prompts, and windows.\n- Approvals: approve or deny Orchestrator BETA requests.\n- Settings: provider detection, Orchestrator BETA MCP registration, and state paths.\n- Doctor: setup and environment health checks only.\n- Reference: compact in-app map of the TUI, CLI, MCP tools, and roles.\n- Credits: package metadata, stack, providers, license, and repository.\n\nThe TUI is visible-menu first: arrows navigate, Enter selects, and Esc\u002FBackspace goes back. Text fields accept normal typing, picker fields use Left\u002FRight, and Welcome also accepts `q` to quit. There are no slash commands, Tab-driven focus panes, hidden command palette, or embedded terminal.\n\n## Visual Design\n\n- The Welcome page animates the blue block logo and shows the duck hero variant.\n- The Runs page uses a static sectioned list to reduce flicker during normal use.\n- The mascot is an in-house chafa-rendered blocky duck (src\u002Fbrand\u002Fassets\u002Fduck.svg).\n- Layouts adapt by page and terminal width instead of forcing one three-pane layout everywhere.\n- The design reference is Claude Code style terminal UX: simple visible controls, compact status, and low redraw noise.\n\n## CLI\n\n```sh\nreevesagents                 # open TUI\nreevesagents spawn [spec...] # start a low-permission multi-terminal spawner run\nreevesagents context         # show caller role, current run, and controls\nreevesagents runs            # list runs\nreevesagents open \u003Cid>       # open a run's reeves window or a terminal\u002Fagent window\nreevesagents peek \u003Cterminal-or-agent-id> # print recent terminal\u002Fagent output\nreevesagents stop \u003Crun-id>   # stop one run, requires --yes or ALLOW_DESTRUCTIVE=1\nreevesagents kill \u003Cterminal-or-agent-id> # close one terminal\u002Fworker, requires --yes or ALLOW_DESTRUCTIVE=1\nreevesagents doctor          # setup checks\nreevesagents orchestrator setup # BETA: register MCP configs\nreevesagents mcp             # BETA: start MCP server over stdio\nreevesagents call \u003Ctool>     # BETA: call any MCP tool from the CLI with JSON args\n```\n\n`call` gives the CLI the same control plane as MCP without adding a separate implementation. Arguments can be inline JSON, stdin JSON, or `--file \u003Cpath>`:\n\n```sh\nreevesagents call context\nreevesagents call get_run '{\"run_id\":\"\u003Crun-id>\"}'\nprintf '%s' '{\"run_id\":\"\u003Crun-id>\"}' | reevesagents call tree\nreevesagents call spawn_worker --caller \u003Croot-agent-id> --file worker.json\n```\n\n## MCP Tools (BETA)\n\n`reevesagents mcp` exposes the Orchestrator BETA control plane. Spawner terminals do not receive MCP environment variables and are not MCP callers.\n\n```text\nstart_run\nlist_runs\ncontext\nlist_agents\ntree\nget_run\nopen_reeves\nopen_agent\npeek\nwait\nsend_text\nsend_key\ninterrupt\nspawn_worker\nkill_agent\nstop_run\nupdate_task\nsend_message\ncheck_messages\nrequest_approval\ncheck_approval\nlist_approvals\nresolve_approval\npoll_approval\nget_inbox\ndoctor\n```\n\nCaller role is inferred from environment:\n\n- no `REEVES_SESSION_ID` or `REEVES_AGENT_ID`: external operator\n- root agent: knows its current run, controls workers, spawns workers, stops the run, and resolves approvals\n- worker agent: knows its current run, updates itself, checks messages, requests approval, and opens Reeves\n\nSee [docs\u002Fmcp-tools.md](docs\u002Fmcp-tools.md) for caller roles, tool groups, and agent usage rules.\n\n## State Layout\n\n```text\n~\u002F.reeves\u002F\n  config.json\n  presets\u002F\n  runs\u002F\n    \u003Crun-id>\u002F\n      run.json\n      agents\u002F\n        \u003Cagent-id>.json\n      approvals\u002F\n        \u003Capproval-id>.json\n      mcp\u002F\n        \u003Cagent-id>-claude-mcp.json\n```\n\n`REEVES_REGISTRY` can point at an isolated state root for tests and smoke runs.\n\n## MCP Config Paths (BETA)\n\n`reevesagents orchestrator setup` writes only the ReevesAgents MCP entry:\n\n- Claude Code: `~\u002F.claude\u002Fsettings.json`\n- Codex CLI: `~\u002F.codex\u002Fconfig.toml`\n- OpenCode CLI: `~\u002F.config\u002Fopencode\u002Fopencode.json`\n- Hermes: `~\u002F.hermes\u002Fconfig.yaml`\n\nNo API keys or provider credentials are written.\n\n## Development\n\n```sh\npnpm install\npnpm typecheck\npnpm lint\npnpm test\npnpm build\n```\n\nPortable verification:\n\n```sh\npnpm verify\n```\n\nReal tmux verification:\n\n```sh\npnpm verify:real\n```\n\nOpt-in real-provider approval drill:\n\n```sh\npnpm smoke:approval-real\nREEVES_REAL_PROVIDER=codex pnpm smoke:approval-real\n```\n\nThe default real approval drill targets Claude Code when installed. Use `REEVES_REAL_PROVIDER` to try another provider.\n\nSee [docs\u002Ftesting.md](docs\u002Ftesting.md) for the isolated test matrix and manual TUI check.\n\nSee [docs\u002Fuse-cases.md](docs\u002Fuse-cases.md) for the product surface map used for careful refactors.\n\nSee [docs\u002Frelease-readiness.md](docs\u002Frelease-readiness.md) for the first public release checklist.\n\nFor future-agent handoff, start with [docs\u002Fagent-brief.md](docs\u002Fagent-brief.md).\n\nSee [docs\u002Fimplementation-report.md](docs\u002Fimplementation-report.md) for the end-to-end v1 implementation notes, removed legacy surfaces, UI changes, and verification record.\n\n## Status\n\nThe TUI was redesigned in May 2026 with Spawner mode as the default low-permission path, Orchestrator BETA for MCP-connected agent teams, a unified color system, responsive frame component, focused pages, auto-cleanup of ended runs, and a visible-menu interaction model. The current source of truth is this README plus [REEVESAGENTS_DESIGN.md](REEVESAGENTS_DESIGN.md).\n\n- Version: `0.9.0`\n- npm package: not published yet\n- Homebrew formula: not available yet\n\n## License\n\nApache 2.0\n","ReevesAgents 是一个免费且开源的AI命令行代理工作区管理器，无需API密钥。它集成了TUI、WebUI、MCP和CLI功能于一体，支持两种模式：默认的Spawner模式可启动多个独立的提供商CLI终端供人类协调；Beta阶段的Orchestrator模式则允许通过MCP工具进行根\u002F工作代理间的连接与协作。项目使用TypeScript编写，适合需要在同一tmux会话中管理和协调多个AI CLI代理的开发者或团队使用。每个运行实例都在本地状态中被追踪，并在自己的tmux窗口中打开，确保了界面整洁的同时提供了强大的多代理协同能力。此外，ReevesAgents不存储提供者凭证也不嵌入终端模拟器，保证了使用的安全性和灵活性。",2,"2026-06-11 03:54:48","CREATED_QUERY"]