By kazukinagata
An async task coordination protocol for autonomous AI agents. Agents discover, claim, execute, and hand off tasks through a shared board.
Generates Acceptance Criteria and Execution Plans for code/technical tasks by exploring the codebase. Returns structured text to the caller.
Generates Acceptance Criteria and Execution Plans for non-code knowledge work tasks (marketing, operations, research, coordination, etc.). Uses domain-specific templates and progressive clarification.
Executes a single development task autonomously. Use when the task-agent skill delegates a Ready task for execution. Reads task description and acceptance criteria, plans implementation, writes code, runs tests.
Reviews a task spec and judges whether a new colleague could reproduce the issuer's intent without asking back. Returns PASS / NEEDS_REFINEMENT / REJECT with per-axis findings and concrete suggested fixes. Spawned by the reviewing-quality shared skill — not invoked directly by user-invocable skills.
Shared logic for resetting fields when assigning a task to another person. Enforces standard field constraints so the recipient can configure their own execution environment. Called by managing-tasks, delegating-tasks, and ingesting-messages.
Shared session bootstrap — detects provider, resolves identity, and populates headless_config. Called by all user-invocable skills at startup.
Delegates a task to another organization member by updating Assignee, resetting executor fields, and recording delegation history in Context. Use this skill whenever the user wants to hand off, transfer, reassign, or give a task to someone else — even if they don't say "delegate" explicitly. Triggers on: "delegate task", "assign to", "transfer task", "reassign", "hand off", "give to", "pass to another person".
Detects the active data source provider and retrieves configuration (database IDs, constants). Internal shared skill — not for direct user invocation.
Orchestrates autonomous task execution via current session, tmux parallel, or Scheduled Tasks. Fetches ready tasks, validates working directories, and dispatches to the chosen execution mode. Use this skill whenever the user wants to do, run, process, start, dispatch, or launch tasks — including parallel execution, batch processing, or working on the next ready task. Explicit execution requests that name a task ID or title (e.g. "execute the X task", "run task ID …") enter through this skill so the Ready-state transition runs through the standard quality gate. Triggers on: "do the next task", "process tasks", "execute tasks", "ready tasks", "run tasks", "start tasks", "dispatch", "launch tasks", "work on next task", "parallel execution", "batch process".
Uses power tools
Uses Bash, Write, or Edit tools
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.
An async task coordination protocol for autonomous AI agents.
Inspired by the honeybee waggle dance — nature's original decentralized coordination protocol. A bee returns to the hive and tells others where to work. No central controller needed.
Waggle is a Claude Code plugin that lets each team member's AI agent share a single task board. Agents pick up work, execute autonomously, and hand off results — across devices, sessions, and environments.
A human creates a task on their laptop. Another member's agent on Claude Desktop picks it up. A third agent running in Cowork executes it overnight. The board is the only shared state.
AI agents today have several ways to collaborate. Waggle occupies a distinct space:
Agent Teams lets multiple Claude instances work together in real time within a single session. When the session ends, the collaboration ends. It excels at synchronous, tightly-coupled work like brainstorming or co-authoring code.
Agent Memory tools (Mem0, Claude-Mem, Engram, etc.) persist knowledge and context across sessions. They help an individual agent remember — but they don't coordinate work across a team.
Waggle coordinates work across team members and their agents. Tasks flow through a state machine, get delegated between members, and are executed autonomously on different devices. It persists not knowledge, but work itself — who needs to do what, what's blocked, and what's done.
| Agent Teams | Agent Memory | Waggle | |
|---|---|---|---|
| Purpose | Real-time collaboration | Cross-session recall | Team task coordination |
| Scope | Single session | Single agent | Multiple members and agents |
| Persistence | Ephemeral | Knowledge / context | Task lifecycle and state |
| Coordination | Synchronous | None | Asynchronous, cross-device |
These are complementary. Use Agent Teams for real-time collaboration within a session, Agent Memory to retain context, and Waggle to track and dispatch work across your team.
Add the marketplace and install:
/plugin marketplace add kazukinagata/waggle
/plugin install waggle@kazukinagata-waggle
Or browse interactively — run /plugin, go to Discover, and select waggle.
/setting-up-tasks
Choose your provider:
"add a task to research competitor pricing"
"what's next?"
"execute tasks"
"show kanban"
"delegate API task to Alice"
Waggle defines 15 core fields and 9 extended fields with a 6-state machine that any backend can implement. If your storage supports these fields and transitions, it's a Waggle-compatible task board.
Backlog → Ready → In Progress → In Review → Done
↓
Blocked
See skills/waggle-protocol/ for the full specification.
| Provider | Use Case | Status |
|---|---|---|
| SQLite | Local, instant, zero-setup | Available |
| Notion | Team collaboration via Notion workspace | Available |
| Turso | Remote SQLite, multi-agent sync | Available |
The provider abstraction means you can add your own backend. See skills/provider-contract/ for the interface contract.
http://localhost:3456npx claudepluginhub kazukinagata/waggle --plugin waggle確定申告を自動化する Claude Code Plugin。会社員+副業(事業所得・青色申告)の所得税・消費税確定申告をエンドツーエンドで支援。
Notion provider for waggle. Enables team task management via Notion workspace.
SQLite provider for waggle. Local zero-config task management.
Turso provider for waggle. Remote SQLite for multi-agent coordination.
Standalone plugin that registers UserPromptExpansion. CLI should accept it; Cowork validation should reject the whole plugin (research section 2.3).
Solo entrepreneur workspace with Mission Control task management, Eisenhower matrix, and AI agent team operations
Context hub for agentic workflows - manage projects, tasks, sessions, and multi-agent coordination
Manage and decompose work into feature-level tasks on the agtx kanban board
Claude Code integration for MCP Task Orchestrator — schema-aware context, note-driven workflow
Persistent Kanban board workflow with TodoWrite sync and SessionStart resume
AI-native task management system. Track progress, coordinate multi-agent work, manage hierarchical tasks with dependencies, sprints, acceptance criteria, and a web dashboard.