[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-942":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":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":25,"hasPages":23,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":31,"readmeContent":32,"aiSummary":33,"trendingCount":16,"starSnapshotCount":16,"syncStatus":14,"lastSyncTime":34,"discoverSource":35},942,"wanman","chekusu\u002Fwanman","chekusu","wanman is an open-source agent matrix runtime inspired by Japanese one-man trains. It lets human users step back into an observer role while local agent runtimes coordinate autonomous multi-agent workflows, task execution, and artifacts.","http:\u002F\u002Fwanman.ai",null,"TypeScript",635,102,2,6,0,15,23,101,45,10.04,"Apache License 2.0",false,"main",true,[27,28,29,30],"agent","claude-code","codex","llm","2026-06-12 02:00:21","# wanman\n\n**English** | [中文](README.zh.md) | [日本語](README.ja.md)\n\nAgent Matrix framework — run a supervised network of Claude Code or Codex agents that collaborate on your machine.\n\nwanman is an open-source local-mode agent matrix framework. Runs a supervised network of Claude Code or Codex agents on your machine, coordinated through a JSON-RPC supervisor.\n\n## About wanman\n\nThe name wanman comes from the Japanese [ワンマン電車 \u002F one-man train](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FOne-person_operation): a train operated by one driver without a conductor. wanman's design goal is similar in spirit: the human user steps back into an observer role and watches the agent matrix run automatically from every angle.\n\n[wanman.ai](https:\u002F\u002Fwanman.ai\u002F) provides the fully automated 24\u002F7 sandbox edition of wanman, running for free on [Sandbank Cloud](https:\u002F\u002Fcloud.sandbank.dev\u002F)'s sandbox cloud.\n\n## What it does\n\n- Coordinates multiple agents (CEO, dev, devops, marketing, feedback, etc.) through an async message bus with steer\u002Ffollow-up priorities.\n- Runs each agent as a real Claude Code or Codex CLI subprocess — you bring your own CLI auth, wanman orchestrates spawning, prompting, and lifecycle.\n- Isolates every agent in a per-agent worktree and per-agent `$HOME`, so agents never mutate your dirty checkout or shell profile.\n- Is CLI-first: everything is scriptable, observable, and reproducible through `wanman` commands and a JSON-RPC supervisor.\n\n[wanman.ai](https:\u002F\u002Fwanman.ai\u002F) adds hosted-only capabilities:\n- Isolates each agent runtime group in its own sandbox environment, supporting large-scale, high-concurrency task execution.\n- Dynamically configures agent roles, including automatically extracting roles from high-quality agent role catalogs on the internet.\n- Supports dynamic skill self-evolution.\n- Supports db9-powered global search and story retrieval.\n\n## Quickstart\n\n```bash\n# Prerequisites: Node 20+, pnpm 9+, git, a logged-in Claude Code or Codex CLI.\ngit clone git@github.com:chekusu\u002Fwanman.git wanman.dev\ncd wanman.dev\npnpm install\npnpm build\n\n# Run from source; no npm package publish is required.\npnpm --filter @wanman\u002Fcli exec wanman takeover \u002Fpath\u002Fto\u002Fany\u002Fgit\u002Frepo\n```\n\nIf you want a single-file CLI bundle instead:\n\n```bash\npnpm --filter @wanman\u002Fcli standalone\nnode packages\u002Fcli\u002Fdist\u002Fwanman.mjs takeover \u002Fpath\u002Fto\u002Fany\u002Fgit\u002Frepo\n```\n\nIf `wanman` is already on your `PATH`, you can also run `wanman takeover .` from inside the target repository.\n\nSee [`docs\u002Fquickstart.md`](docs\u002Fquickstart.md) for the full walkthrough.\n\n## CLI commands\n\n| Command | What it does |\n|---------|--------------|\n| `wanman send \u003Cagent> \u003Cmsg>` | Send a message to an agent (`--steer` interrupts the target). |\n| `wanman recv [--agent \u003Cname>]` | Receive and mark pending messages as delivered. |\n| `wanman agents` | List registered agents and their current states. |\n| `wanman context get` \u002F `context set` | Read or write shared key\u002Fvalue context. |\n| `wanman escalate \u003Cmsg>` | Escalate to the CEO agent. |\n| `wanman task …` | Manage the task pool: `create`, `list`, `get`, `update`, `done`. Supports `--after` dependencies. |\n| `wanman initiative …` | Manage long-lived initiatives: `create`, `list`, `get`, `update`. |\n| `wanman capsule …` | Manage change capsules: `create`, `list`, `mine`, `get`, `update`. |\n| `wanman artifact …` | Store and retrieve structured artifacts: `put`, `list`, `get`. |\n| `wanman hypothesis …` | Track hypotheses with status transitions: `create`, `list`, `update`. |\n| `wanman watch` | Live-stream supervisor and agent activity. |\n| `wanman run \u003Cgoal>` | Start a matrix against a one-shot goal. |\n| `wanman takeover \u003Cpath>` | Take over an existing git repo with the full agent matrix. |\n| `wanman skill:check [path]` | Validate that skill docs reference only real CLI commands. |\n\nRun `wanman --help` for the full, current list.\n\n## Architecture\n\n```\n+----------------+          +--------------------+          +-----------------+\n|  wanman CLI    |  JSON    |  Supervisor        |  spawn   |  Agent process  |\n|  (host shell)  | ---RPC-->|  (local process)   | -------> |  (Claude\u002FCodex) |\n|                |  \u002Frpc    |  message\u002Fcontext\u002F  |          |  per-agent $HOME|\n|                |          |  task\u002Fartifact     |          |  per-agent wt   |\n+----------------+          +--------------------+          +-----------------+\n                                    |                              |\n                                    v                              v\n                           +--------------------+        +-----------------+\n                           |  files + SQLite    |        |  worktree       |\n                           +--------------------+        +-----------------+\n```\n\n- `wanman \u003Csubcommand>` speaks JSON-RPC 2.0 to a Supervisor process.\n- The Supervisor owns the message store, context store, task pool, artifact store, and spawns one child process per agent.\n- Each agent child is a local Claude Code or Codex subprocess bound to a per-agent worktree and isolated `$HOME`.\n\nDeep dive: [`docs\u002Farchitecture.md`](docs\u002Farchitecture.md).\n\n## Configuration\n\n| Env var | Meaning |\n|---------|---------|\n| `WANMAN_URL` | Supervisor HTTP URL for the CLI (default `http:\u002F\u002Flocalhost:3120`). |\n| `WANMAN_AGENT_NAME` | Identifies the current agent; used as default sender\u002Freceiver inside agent processes. |\n| `WANMAN_RUNTIME` | `claude` (default) or `codex` — selects the per-agent CLI adapter. |\n| `WANMAN_MODEL`, `WANMAN_CODEX_MODEL`, `WANMAN_CODEX_REASONING_EFFORT` | Per-runtime model overrides. |\n| `WANMAN_CODEX_FAST` | When set, biases the Codex adapter toward lower-latency defaults. |\n| `WANMAN_SKILL_SNAPSHOTS_DIR` | Override where the runtime materializes skill-activation snapshots (default: sibling of the shared-skills dir, falling back to `$TMPDIR\u002Fwanman-skill-snapshots`). |\n\nAn optional `@sandbank.dev\u002Fdb9` brain adapter can be attached for cross-run memory — see [`docs\u002Farchitecture.md`](docs\u002Farchitecture.md#brain--persistence).\n\n## Agent configs\n\nAgent definitions live in a single JSON file:\n\n```json\n{\n  \"agents\": [\n    { \"name\": \"echo\", \"lifecycle\": \"24\u002F7\", \"model\": \"standard\", \"systemPrompt\": \"...\" },\n    { \"name\": \"ping\", \"lifecycle\": \"on-demand\", \"model\": \"standard\", \"systemPrompt\": \"...\" }\n  ],\n  \"dbPath\": \".wanman\u002Fwanman.db\",\n  \"port\": 3120,\n  \"workspaceRoot\": \".wanman\u002Fagents\"\n}\n```\n\nEach agent entry has:\n- `name` — unique identifier used on the message bus.\n- `lifecycle` — `24\u002F7` (continuous respawn loop), `on-demand` (idle until triggered), or `idle_cached` (idle until triggered, but the prior Claude `session_id` is preserved across triggers via `claude --resume` so context survives idle periods). **`idle_cached` is Claude-only**: pairing it with `runtime: codex` (or `WANMAN_RUNTIME=codex`) is rejected at startup since Codex has no equivalent resume mechanism in this runtime.\n- `model` — usually an abstract tier (`high` or `standard`); the runtime adapter maps it to Claude or Codex defaults, with environment overrides available.\n- `systemPrompt` — baked-in persona\u002Fmission; agents also auto-discover shared skill files at `~\u002F.claude\u002Fskills\u002F`.\n- Optional `cron`, `events`, and `tools` fields — see the architecture doc for the full schema.\n\n## Testing\n\n```bash\npnpm typecheck\npnpm test\npnpm exec vitest run --coverage --coverage.reporter=text-summary --coverage.reporter=json-summary --coverage.exclude='**\u002Fdist\u002F**'\n```\n\nThe current coverage target is at least 90% line coverage. The latest verified local run reports `Lines: 90.17%`; the machine-readable summary is written to `coverage\u002Fcoverage-summary.json`.\n\n## Project structure\n\n```\nwanman.dev\u002F\n  packages\u002F\n    cli\u002F                 wanman CLI (send\u002Frecv\u002Ftask\u002Fartifact\u002Frun\u002Ftakeover\u002F...)\n    core\u002F                Shared types, JSON-RPC protocol, skills (core\u002Fskills\u002F)\n    host-sdk\u002F            Host-side SDK for embedding wanman into other tools\n    runtime\u002F             Supervisor, agent process manager, SQLite stores, adapters\n  docs\u002F                  Architecture and quickstart guides\n```\n\nShared skills shipped today (`packages\u002Fcore\u002Fskills\u002F`):\n- `artifact-naming`, `artifact-quality` — conventions for agent-produced artifacts.\n- `cross-validation` — CEO consistency checks across agent outputs.\n- `research-methodology` — market\u002Fdata research methodology.\n- `wanman-cli` — CLI command reference consumed by agents at runtime.\n- `workspace-conventions` — file-system conventions inside agent workspaces.\n\n## Further reading\n\n- [Quickstart](docs\u002Fquickstart.md) — first-run walkthrough against any git repo.\n- [Architecture](docs\u002Farchitecture.md) — agent lifecycle, JSON-RPC, stores, adapters.\n- [Contributing](CONTRIBUTING.md) — tests, typecheck, commit conventions.\n\n## License\n\nApache-2.0 — see [LICENSE](LICENSE).\n","wanman 是一个受日本单人列车启发的开源代理矩阵运行时，旨在让人类用户退居观察者角色，同时本地代理运行时协调自主多代理工作流、任务执行和制品。项目采用 TypeScript 编写，通过 JSON-RPC 监督器协调 Claude Code 或 Codex 代理网络。每个代理在独立的工作树和 HOME 环境中运行，确保不会影响用户的代码库或 shell 配置。wanman 适合需要自动处理复杂任务分配和协作的开发环境，如软件开发、运维、市场营销等场景。","2026-06-11 02:40:22","CREATED_QUERY"]