A Markdown-only agent team workflow for Claude Code and Codex.
\n
\n\n \n\n\n\n## Why\n\nA **Mission** is a bounded objective. A **Squad Lead** decomposes it, names a squad, and publishes tasks to a kanban. Background Agents claim tasks, do the work, and write results back. Every handoff is a Markdown edit you can open, diff, and review.\n\nThe kanban file *is* the coordination protocol. If you can read `kanban.md`, you know what every agent is doing and why.\n\nThis is a standalone plugin for **Claude Code** and **Codex**. It does not require a Mexus server, panes, a Mission Inbox, or external A2A messaging — coordination happens entirely through the `agent-team/` Markdown files.\n\n## How It Works\n\nCoordination lives entirely in one directory of your project:\n\n```text\nagent-team/\n├── mission-workflow.md the reusable collaboration rules\n├── agents.md repository-level reusable agent roster\n└── missions//\n ├── mission.md goal, lifecycle, constraints\n ├── agents.md this Mission's squad and prompts\n ├── kanban.md task source of truth — the protocol\n ├── roundtable.md shared decisions\n ├── squad-lead.md Squad Lead coordination log\n └── artifacts/ long-form deliverables (research, decisions, docs)\n```\n\n- **Squad Lead** is a fixed coordination role — it plans, dispatches, sequences, and accepts. It never writes another agent's results.\n- **Mission Agents** are background workers, named at random from the 72 spirits of the Ars Goetia. Each claims one scoped task, self-tests it, and fills in its own kanban block.\n- **Publisher acceptance**: whoever publishes a task (`From`) reviews and accepts it. Squad Lead does mission-level acceptance on top.\n\nStarting a new Mission archives the previous active one. Agents communicate only through the kanban file.\n\n## Artifacts\n\nLong-form deliverables that are not source code — research reports, market scans, decision rationales, design memos — live as Markdown files under `agent-team/missions//artifacts/`. Each file starts with a YAML frontmatter block (`title`, `author`, `date`, optional `related` and `tags`) so the board can list and render it.\n\nThree rules keep artifacts from being skipped:\n\n- **Research and document tasks** must name the artifact path in their `Acceptance` field. The executor writes the file and lists the path in `Files:`.\n- **RoundTable proposals that pass review** must produce a decision-record artifact (`artifacts/-rt-.md`) before they are considered closed.\n- **Mission-level acceptance** checks both of the above before Squad Lead declares the Mission complete.\n\nThe board's **Artifacts tab** renders each file as HTML and shows author, date, related refs, and tags. See `references/artifact-template.md` for the file skeleton and `agent-team/mission-workflow.md` (`## Artifacts`) for the full rules.\n\n## Commands\n\nPlugin skills are namespaced by the plugin. This plugin uses the `mexus-team` namespace.\n\n| Command | When | What it does |\n| :--- | :--- | :--- |\n| [`/mexus-team:mission`](skills/mission/SKILL.md) `\"\"` | Starting a new piece of work | Creates a Mission, archives the old one, plans the squad and kanban, starts the board, and dispatches the first background Agents. |\n| [`/mexus-team:continue`](skills/continue/SKILL.md) | Resuming a Mission in a later session | Reads the existing kanban and restarts outstanding tasks. No re-planning, no archiving. |\n| [`/mexus-team:roundtable`](skills/roundtable/SKILL.md) `\"\"` | A decision affects multiple agents | Opens a RoundTable proposal for a shared decision. |\n| [`/mexus-team:board`](skills/board/SKILL.md) | Watching a Mission run | Starts a local read-only Web board over `agent-team/`. |\n| [`/mexus-team:status`](skills/status/SKILL.md) | A quick terminal check | Prints a concise Mission and task summary. |\n\nWithin one `/mexus-team:mission` session, Squad Lead dispatches follow-up tasks on its own — you do not run a command for that. `/mexus-team:continue` exists only to pick a Mission back up after the original session has ended.\n\n## Install\n\nIn Claude Code, add the marketplace from the GitHub repo and install the bundle:\n\n```text\n/plugin marketplace add yofine/mexus-agent-team\n/plugin install mexus-team@mexus-team\n/reload-plugins\n```\n\n`mexus-team@mexus-team` is `@`. Both are defined in `.claude-plugin/marketplace.json`; the GitHub repo slug (`mexus-agent-team`) is not the install handle.\n\nThe bundle registers all five commands under the `mexus-team:` namespace. The marketplace also carries per-skill entries such as `mexus-team-mission@mexus-team` for future single-skill installs; install the bundle for now.\n\nTo update after the upstream changes:\n\n```text\n/plugin uninstall mexus-team\n/plugin install mexus-team@mexus-team\n/reload-plugins\n```\n\n## Workflow\n\nA typical Mission, start to finish:\n\n1. **Create and run** — `/mexus-team:mission \"Implement the feature or fix\"`. This plans the squad, publishes kanban tasks, starts the board, and dispatches the first Agents. Squad Lead keeps coordinating follow-up work in the same session.\n2. **Watch** — open the board separately with `/mexus-team:board` if you want a live view. It shows Mission Plan, Kanban, Agents, Squad Lead Log, and RoundTable.\n3. **Decide** — `/mexus-team:roundtable \"Choose the API boundary\"` when a choice spans multiple agents or shared interfaces.\n4. **Check** — `/mexus-team:status` for a quick terminal summary.\n5. **Resume** — `/mexus-team:continue` in a new session to keep an unfinished Mission moving.\n\n## Board\n\nThe board command starts a local Vite app and an API server. By default it binds to `0.0.0.0`, so it can be opened from another device on your network. If a board is already running for the project, it is reused and its URL reprinted instead of starting a second one.\n\n```text\n/mexus-team:board --host 0.0.0.0\n/mexus-team:board --host 0.0.0.0 --public-host \n```\n\nThe board is read-only and runs in the foreground — stop it with `Ctrl+C` in its terminal.\n\n## Roster Maintenance\n\n`agent-team/agents.md` is the repository-level roster of reusable Agents. It is updated from accepted Mission work, not edited as scratch space — the active Mission's `agents.md` and accepted `kanban.md` tasks are the source for durable module ownership, known strengths, and work history.\n\n## Design Notes\n\n- The whole protocol is Markdown in the repo. Coordination is a file edit, review is a diff, and the history is in `git log`.\n- `Squad Lead` is the single fixed coordination role; execution Agents are named at random from the Ars Goetia set, so a name is a handle and not a task label.\n- Commands have non-overlapping jobs: `mission` only creates, `continue` only resumes, the rest only observe.\n\n## Verify Locally\n\n```bash\nnode tests/standalone-smoke.test.mjs\n```\n\nTry it against a temp project:\n\n```bash\nmkdir -p /tmp/mexus-agent-team-demo\ncd /tmp/mexus-agent-team-demo\n\nnode /absolute/path/to/mexus-agent-team/scripts/start-mission.mjs \\\n --root \"$PWD\" --name demo-mission --request \"Build a small demo feature\"\n\nnode /absolute/path/to/mexus-agent-team/scripts/status.mjs --root \"$PWD\" --name demo-mission\nnode /absolute/path/to/mexus-agent-team/scripts/start-board.mjs --root \"$PWD\" --host 0.0.0.0\n```\n\n## License\n\nApache-2.0. Use it and contribute back.\n",
"bytes": 7659,
"sha": "df184d560400201cb46baa25732b76e6cf9a86698667ef6ec0615ba0ba4c69c0",
"repo_slug": "yofine/mexus-agent-team",
"fonte": "repo",
"truncated": false,
"api": "https://agentalog.com/api/listings/plg_yofine_mexus_agent_team_mexus_agent_team_6f7e8752/readme"
}