agiflow — плагин для Claude Code: что делает и как установить
Связывает агента с AgiFlow через MCP для ведения задач, планирования фич, проведения груминга и контроля выполнения прямо из чата.
MIT ★ 1 596 обновлён 2026-10-29
/plugin marketplace add hashgraph-online/awesome-codex-plugins
/plugin install agiflow@awesome-codex-pluginsЧто делает agiflow
Плагин добавляет в Claude Code и Codex поддержку процессов AgiFlow. Внутри зашиты десять специализированных скиллов под разные фазы разработки: декомпозиция требований (project-plan), уточнение деталей (refine-task), приоритизация бэклога (backlog-grooming), исполнение задач (run-task, run-work) и проверка результатов по критериям приёмки (review-work). Вызывать их вручную обычно не нужно — агент сам активирует нужный сценарий по смыслу запроса.
Интеграция работает через удалённый MCP-сервер https://agiflow.io/api/v1/mcp-app с авторизацией по OAuth. Через инструменты сервера агент создаёт и обновляет проекты, группирует задачи в рабочие единицы, оставляет комментарии и распределяет работу между участниками.
Без авторизации в AgiFlow через OAuth удалённый сервер отклонит запросы, поэтому сессию нужно настроить до запуска рабочих пайплайнов.
Какие задачи решает
- разбить требования к фиче на рабочие подзадачи
- провести груминг бэклога и перевести готовые тикеты в работу
- запустить сквозное выполнение задачи или рабочего пакета
- собрать статус проекта и блокирующие проблемы для дейли
Примеры запросов агенту
- Спланируй фичу: настройки уведомлений для каждого пользователя
- Проведи груминг бэклога и переведи готовые задачи в статус Todo
- Собери отчёт daily standup по текущему проекту
Как установить agiflow
Через маркетплейс плагинов
/plugin marketplace add hashgraph-online/awesome-codex-plugins/plugin install agiflow@awesome-codex-pluginsПо шагам, с командами для каждой платформы: Плагины Claude Code: маркетплейсы, установка и как собрать свой
Вопросы об agiflow
Как установить agiflow в Claude Code?
Выполните по очереди две команды в терминале Claude Code: сначала /plugin marketplace add hashgraph-online/awesome-codex-plugins, затем /plugin install agiflow@awesome-codex-plugins.
Что делает плагин agiflow?
Он подключает клиента к серверу AgiFlow через протокол MCP и даёт агенту готовые сценарии для планирования, трекинга задач и оркестрации разработки.
Как проверить подключение MCP-сервера после установки?
Внутри интерфейса Claude Code введите команду /mcp, чтобы увидеть статус соединения с эндпоинтом AgiFlow.
Исходный текст
ai-plugin · MIT
# AgiFlow AI Plugin Official AgiFlow plugin for AI clients. Drive AgiFlow project management, including planning, grooming, execution, and review, directly from your AI tool. Works with **ChatGPT**, **Claude Code**, **Codex**, **Cursor**, **Antigravity**, and **Gemini CLI**. ## Installation This repo is a self-contained, multi-client plugin bundle. Until it is published to each client's marketplace, load it as a local plugin directory. ### Claude Code Install from the AgiFlow marketplace (Claude Code 2.1.275 or later): ``` /plugin install agiflow --marketplace AgiFlow/ai-plugin ``` For local development, load this checkout directly: ```bash claude --plugin-dir ./ai-plugin ``` The bundled `.mcp.json` wires the AgiFlow MCP server automatically. Use `/mcp` inside Claude Code to check the connection. ### Antigravity (Google) Place the plugin folder in one of Antigravity's plugin locations, then restart: ```bash # Workspace-level (this project only) mkdir -p .agents/plugins && cp -R /path/to/agiflow-ai-plugin .agents/plugins/ # Global (all workspaces) mkdir -p ~/.gemini/config/plugins && cp -R /path/to/agiflow-ai-plugin ~/.gemini/config/plugins/ ``` Antigravity reads the root `plugin.json` marker, the `skills/`, and `mcp_config.json` automatically. ### Cursor Add manually in **Cursor Settings → MCP / Plugins**, pointing at this folder. Cursor's stable surface is MCP config. The bundled `.mcp.json` provides it. ### Codex Add the AgiFlow plugin marketplace, then install the plugin from that marketplace: ```bash codex plugin marketplace add AgiFlow/ai-plugin codex plugin add agiflow@agiflow ``` For local development, point Codex at this checkout as a marketplace root: ```bash codex plugin marketplace add ./agiflow-ai-plugin codex plugin add agiflow@agiflow ``` ### Gemini CLI ```bash gemini extensions install <your-remote>/agiflow-ai-plugin ``` The bundled `gemini-extension.json` connects the AgiFlow MCP server via `mcp-remote`. ### ChatGPT OpenAI Platform plugins are submitted from the production AgiFlow MCP server. This public repository provides the reusable Agent Skills used by that plugin, but does not contain Platform dashboard IDs, submission evidence, reviewer credentials, or other private operational data. The ChatGPT-safe workflow set is: - `getting-started` - `project-plan` - `refine-task` - `backlog-grooming` - `daily-standup` - `triage` The coding-agent workflows `orchestrate`, `run-task`, `run-work`, and `review-work` remain available to coding clients but are not part of the ChatGPT submission bundle. ## How to develop ```bash git clone https://github.com/AgiFlow/ai-plugin.git claude --plugin-dir ./ai-plugin ``` - Add new workflow instructions under `skills/<name>/SKILL.md`. - Keep shared guidance in `references/` (e.g. `references/agiflow-agents.md`). - See `references/plugin-types.md` for per-client manifest notes. ## Features This plugin connects to the AgiFlow MCP server (`https://agiflow.io/api/v1/mcp-app`) and exposes AgiFlow tools across these categories: - **Projects**: create, inspect, and update projects and their statuses - **Tasks**: create, list, get, update, reorder, and batch-create tasks - **Work units**: group tasks into deliverable features or epics and track progress - **Workflows**: acquire and release locks and coordinate multi-agent runs - **Members**: list and assign agent members to work - **Comments**: document decisions and progress on tasks - **Vault**: read and set scoped configuration entries ### Bundled skills The plugin ships 10 workflow skills that mirror AgiFlow's scrum pipeline. Your AI client loads them on demand when your request matches their description. You generally do not invoke them by name: | Skill | Phase | Use it to | | --- | --- | --- | | `getting-started` | orient | get coached on where to start and which workflow fits | | `project-plan` | Planning | break requirements into vertical-slice tasks (Planning status) | | `refine-task` | Planning | turn a vague task into an autonomous-ready spec | | `backlog-grooming` | Planning → Todo | verify, prioritize, and promote tasks into work units | | `run-work` | Todo → Done | execute a whole work unit end-to-end in one session | | `run-task` | Todo → Done | execute a single task through to Review | | `review-work` | Review | verify acceptance criteria and file follow-ups | | `triage` | diagnose | classify project issues by severity and recommend actions | | `daily-standup` | report | a read-only pulse of done / in-progress / blocked / next | | `orchestrate` | dispatch | route the highest-priority ready work to agents | Shared guidelines (status model, transitions, tags, work-unit sizing) live in [`references/agiflow-agents.md`](references/agiflow-agents.md). ## Example usage ``` > Plan a feature: add per-user notification preferences > Groom the backlog and promote the ready tasks to Todo > Run task DXX-2 > Execute the checkout work unit end-to-end > Review the auth work unit against its acceptance criteria > Give me a daily standup for this project > Why is this project stuck? > What should an agent pick up next? ``` ## Data handling and external services The declared connector is `https://agiflow.io/api/v1/mcp-app`, authenticated with OAuth. Agiflow can read and store names, email addresses, member profiles, tasks, comments, and artifacts that you provide or authorize it to access. Workspace content is retained beyond 30 days unless removed, subject to the [Privacy policy](https://agiflow.io/privacy-policy). Some workflows can send data beyond the declared Agiflow connector: - **External MCP integrations:** When an allowed external tool is invoked through `task_exec` or `task_exec_code`, Agiflow forwards the tool arguments to the MCP server configured for your workspace and task. Destinations depend on the integrations you enable; there is no fixed third-party server list bundled with this plugin. Arguments can include project content and personal data supplied for the requested operation. The external provider's privacy policy applies. - **Approved code scripts:** `task_exec_code` can call task-allowed MCP tools. An approved project script can also send HTTPS requests, including request bodies, to hosts permitted by its approved environment configuration. One-off code has no direct network access. Approving a script or tool can authorize reads and writes in an external service. - **Coding-client tools:** Coding workflows instruct the AI client to use tools in your environment, including `vibe-lint`, local tests, Git, and pull-request tooling. Source code and project context can be passed to those tools or your configured Git hosting service. Their destinations depend on your client configuration. Gemini CLI downloads and runs `mcp-remote` from npm via `npx`. - **Optional analytics:** When configured by the Agiflow deployment, the backend sends MCP usage events to PostHog. These events can include user, organization, project, task, and session IDs, client information, tool names, timing, success/error information, and submitted feedback. Only send data to integrations and scripts you trust. Review their permissions and data handling before enabling them. Also see the [Terms of Service](https://agiflow.io/terms-and-conditions). ## Self-hosted For a self-hosted AgiFlow instance, point the MCP wiring at your endpoint via the `AGIFLOW_AI_PLUGIN_MCP_URL` environment variable (consumed by `gemini-extension.json`): ```bash export AGIFLOW_AI_PLUGIN_MCP_URL="https://mcp.your-agiflow-instance.com/api/v1/mcp-app" ``` For other clients, edit the server URL in `.mcp.json`, `mcp.json`, and `mcp_config.json`. ## Documentation - AgiFlow: https://agiflow.io - Plugin client compatibility: [`references/plugin-types.md`](references/plugin-types.md) ## License MIT
© AgiFlow · MIT · hashgraph-online/awesome-codex-plugins