By himkt
Orchestrate design doc and research report creation through multi-agent workflows (director, drafter, reviewer) with lifecycle management, multi-round interviews, and TDD execution. Interact with the CAFleet message broker to register agents, send/receive messages, poll inbox, and supervise agent teams with heartbeat and recovery.
Use when the user asks to create a design doc, design document, specification, or technical spec (create workflow → Director/Drafter/Reviewer team); to validate, review, or interview an existing design doc through multi-round Q&A (interview workflow); or to implement or execute a design doc (execute workflow → TDD team). Also the standardized format spec — consult the template and guidelines when editing a design doc. Always invoke this skill and route into the matching workflow, orchestrated as a CAFleet team. Teammates in agent teams load this skill by its name cafleet-design-doc via their backend's skill-loader.
Use when the user asks to research a topic or create a multi-source research report (report workflow → Director/Manager/Researcher team writing to researches/<topic>/); to build a presentation, slide deck, or reading transcript from a report (presentation workflow); to create a chart, plot, graph, figure, or data visualization (visualization reference); or to author slides with the custom Slidev theme (slidev reference). Always invoke this skill and route into the matching workflow, orchestrated as a CAFleet team. Members in agent teams load this skill by its name cafleet-research via their backend's skill-loader.
Interact with the CAFleet message broker and supervise CAFleet agent teams. Use when an agent needs to register, send/receive messages, poll inbox, acknowledge messages, or discover other agents; or when a Director is about to spawn, monitor, health-check, or recover a stalled team of CAFleet members (any `cafleet member create`), which requires the dedicated monitoring member, the heartbeat, and the supervision governance.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
https://github.com/user-attachments/assets/bd2b195a-f3de-4fa3-bcc8-3c6ef9f1016a
Agent Teams reinvented for collaborative coding across multiple coding-agent backends (Claude Code, Codex, and OpenCode), with full code transparency.
claude, codex, and opencode members in one fleet.CAFleet works with three coding agents: claude (Claude Code), codex (OpenAI Codex CLI), and opencode.
Install the plugin in whichever one you use — the broker CLI is shared.
Use your favorite tool to install skills (for example, GitHub CLI gh skill, vercel skills, or the marketplace of each coding agent).
gh skill install himkt/cafleet --agent claude-code
gh skill install himkt/cafleet --agent codex
gh skill install himkt/cafleet --agent opencode
/plugin marketplace add himkt/cafleet
/plugin install cafleet@cafleet
codex plugin marketplace add himkt/cafleet
uv tool install cafleet # or: pip install cafleet
cafleet db init # apply schema migrations (idempotent; rerun after upgrades)
The default database is ~/.local/share/cafleet/cafleet.db. Override with CAFLEET_DATABASE_URL (use an absolute path — SQLAlchemy does not expand ~ in SQLite URLs).
Per-coding-agent config (Claude permissions.allow, Codex config.toml + rules, opencode) lives on the Configure page: https://himkt.github.io/cafleet/get-started/configure/.
Provide the following prompt to Claude Code or Codex to see how it works.
I want to see how cafleet works.
Please create a new team with two members using cafleet and let them ping-pong each other.
After the demonstration, please shutdown the team.
CAFleet provides the builtin skills for Spec Driven Development (SDD). We're using CAFleet to develop CAFleet!
Invoke the cafleet-design-doc skill (create workflow) with a one-line request, e.g.:
I want to create a simple TUI calculator. Please create a design doc using the CAFleet skill.
See your coding-agent's skill documentation for the literal invocation syntax (Claude Code's /skills, codex's /skills, opencode's skill discovery).
You can see the existing design docs on design-docs/, which are actually created by the skills.
CAFleet ships a unified cafleet CLI and an admin WebUI on top of a single-file SQLite database. Fleets partition agents into isolated namespaces; the CLI accesses SQLite directly through a shared broker module, so no HTTP server is required for agent operations. Members spawn as tmux panes running any of the three coding-agent backends, optionally pinned to a specific LLM via cafleet member create --model <m> (e.g. sonnet, gpt-5.4-mini, anthropic/claude-sonnet-4-6, opencode/big-pickle). Full architecture documentation is published at https://himkt.github.io/cafleet/concepts/overview/.
A Director supervises its team on a periodic heartbeat supplied by cafleet monitor — a per-fleet loop the fleet's dedicated monitoring member runs as a background task in its own pane. It spends no model tokens and works on any backend: the monitor schedules the heartbeat, and the monitoring member inspects each due agent and re-engages an idle Director on demand. Spawn the monitoring member first (cafleet member create … --role monitor --model haiku); it launches cafleet monitor start, and stopping that background task or fleet delete ends the loop. See https://himkt.github.io/cafleet/concepts/monitoring/.
Build, test, and project-structure instructions, plus the design-doc-driven contribution flow, are published at https://himkt.github.io/cafleet/get-started/contributing/.
npx claudepluginhub himkt/cafleet --plugin cafleetMulti-agent orchestration for Claude Code. 12 specialized agents working in parallel — planning, building, reviewing, debugging. Plus a Hub for always-alive multi-project sessions controllable from Telegram or Slack.
Orchestrate multi-agent teams for parallel code review, hypothesis-driven debugging, and coordinated feature development using Claude Code's Agent Teams
Multi-agent development orchestration with Claude Code + cmux. Spawn, monitor, and integrate parallel sub-agents visually in terminal panes.
Agent Teams スキルを設計・構築するためのベストプラクティスガイド。サブエージェント定義、SendMessage 通信プロトコル、タスク依存管理、PostToolUse Hook ログ、MCP ツール統合、コンテキストファイル設計を網羅。7つの実績あるチームスキルから抽出したパターン集
Inter-agent communication for Claude Code and Codex CLI sessions via threads and messages
Multi-agent team orchestration for Claude Code. Set up parallel AI agent teams with file-based planning, progress tracking, and role-based collaboration.