Tools to sync AI agent skills, practices, and workflows across repositories.
Skills are authored once under skills/<techno>/<name>/SKILL.md and translated
into each target platform's format (Claude Code, GitHub Copilot, Cursor,
Windsurf).
It ships the maggie command (render skills into repos, bootstrap a local
workspace, run the board and the agents server) — a kanban dashboard and the
server that launches local agents from it (AI Factory), as an
Nx monorepo on pnpm workspaces. The board tracks local agent
sessions for both Claude Code and GitHub Copilot CLI.
Every repository in an organization ends up with its own drifting copy of the
same agent instructions — one for Claude Code, another for Copilot, a third for
Cursor. maggie keeps one reviewed source of truth: write the guidance once,
declare which technologies each repo uses, and let the tool render and open the
pull requests.
Requires Node.js >= 22 and git on your PATH.
npm install -g @linktogo/maggie-board
# See what would be pushed to the repos in your config — no side effects
maggie sync --config repos.json --dry-runThe same install gives you the board and the server (via @linktogo/maggie-server), both
self-contained (the board's front-end ships pre-built, the skills library is bundled):
maggie # the board (http://localhost:4180) and the server (http://127.0.0.1:4318)
maggie --board # the board only (reads ./wk/.maggie/board.json or ./board.json)
maggie --server --repos repos.json # the server only
maggie --repos repos.json --port 5000 --server-port 5100 # options go to the part they belong to
maggie sync --config repos.json # render the skills into the repos
maggie workspace --config repos.json --workspace wk # bootstrap a local workspaceOptions are added after the mode and go to whichever part takes them (see
the board and the server); with both
running, --port is the board's and --server-port the server's, and the board is
pointed at it. --config-repo / --config-file go to both. --board <file> (with a
value) still names the board file; a bare --board picks the mode.
Both answer on the loopback interface only. Without a global install, use
npx -p @linktogo/maggie-board maggie (the server alone: npx @linktogo/maggie-server).
The installed packages bundle the skills library, so the commands work from any
directory. A skills/ folder in the current directory takes precedence, which
is what you want when working from a clone; --skills <dir> overrides both.
To hack on the project itself, work from a checkout:
git clone https://github.com/linktogo/maggie.git
cd maggie
pnpm install
node apps/board/bin/maggie.js sync --config repos.example.json --dry-runThen copy repos.example.json, point it at your own repositories, and drop the
--dry-run when the preview looks right.
pnpm run wk # clone those repos into ./wk and print the command to open them
pnpm start # serve the board (auto-detects wk/.maggie/board.json)pnpm run wk wires Claude Code hooks by default; add -- --agent copilot to
wire GitHub Copilot CLI instead. Sessions from both show up on the same board,
each card labelled with the agent behind it — see
maggie workspace CLI.
New here? Read Configuration next, then Adding a skill.
Full reference lives in docs/:
| Page | What it covers |
|---|---|
| Configuration | The repos.json schema, and the two ways both CLIs resolve it |
| Skills library | Authoring skills, how they map to each target platform |
maggie CLI |
Rendering skills into repos and pushing them |
maggie workspace CLI |
Bootstrapping a workspace, worktrees, status tracking |
| Board dashboard | The kanban dashboard, its server and endpoints |
| CI status | Per-contributor CI badges on the board, and how to enable them |
| Retro-documentation | Folding a repository's specs and plans into one document an agent can read |
| AI Factory | Launching local Claude Code / Copilot agents from the board through an independent server |
| Architecture | Nx layout, module boundaries, testing and coverage gates |
docs/superpowers/ holds the design record — one spec and plan per feature,
kept as history rather than maintained.
skills/ ships a starter set of guidance, grouped by technology and matched
against each repo's technologies list: nestjs, postgres, nextjs,
reactjs, angular, vuejs, nx, firebase, cloudflare.
Add one by creating skills/<techno>/<name>/SKILL.md with YAML frontmatter
(name, description, optional globs) followed by the guidance body. Any
repo whose technologies include <techno> picks it up on the next sync — see
Skills library.
Everything publishes under the single @linktogo scope. The maggie command ships in
@linktogo/maggie-board (the package you install, with the board UI), which runs the agents
server shipped in @linktogo/maggie-server; the libraries are published independently so they
can be reused on their own:
| Package | What it gives you |
|---|---|
@linktogo/maggie-config |
load/validate the repo config from a file or a git repo |
@linktogo/maggie-git |
thin git/gh wrapper |
@linktogo/maggie-renderers |
render a skill for claude/copilot/cursor/windsurf |
@linktogo/maggie-skill-sync |
skill resolution + the sync pipeline |
@linktogo/maggie-workspace-bootstrap |
clone/install, Claude Code & Copilot CLI hooks, board state |
@linktogo/maggie-ci-status |
CI status payloads, validation and state mapping |
@linktogo/maggie-retro-doc |
fold a repo's specs and plans into one agent-facing document |
@linktogo/maggie-agent-runner |
run local agent CLIs on a checkout, with streamed output |
All are released in lockstep on the same version — see
Releasing. The two apps are published too:
@linktogo/maggie-board (the maggie command and the board) and @linktogo/maggie-server.
pnpm test # nx run-many -t test: every lib/app, 100% coverage gate each
pnpm run test:board # apps/board suite only: server (node:test) + front-end (vitest)
pnpm run check:coverage # fails if any project no longer enforces 100% coverageCI runs nx run-many -t lint test build. Because Nx detects the package
manager from the lockfile, keep pnpm-lock.yaml as the only lockfile — a
package-lock.json breaks the Nx project graph. Layout, testing conventions and
the reasons behind them are in Architecture.
Contributions are welcome — especially new skills, which are the easiest way in. Start with CONTRIBUTING.md for the dev setup, the coverage bar, the commit format, and how to add a skill. Participation is governed by our Code of Conduct.
- 🐛 Report a bug
- 💡 Propose a feature or a skill
- 🔒 Security issues: see SECURITY.md — never a public issue
Release notes live in CHANGELOG.md.
The CLIs run git and package-manager commands on your machine: they clone the
repositories named in the config, install their dependencies, and shell out to
gh when --pr is passed. Treat the config file — and any repository you pass
to --config-repo — as trusted input. The board server is a local development
tool with no authentication and should not be exposed to a network. See
SECURITY.md for the full scope and reporting process.
Apache License 2.0 © Linktogo.
Third-party material redistributed in this repository is listed in THIRD_PARTY_NOTICES.md.