Setup logo

Setup

Organization
apache
setup

Install Magpie, configure it for yourself, or adopt it for a repo. Installing touches only this machine; configuring writes gitignored local files; adopting commits a floor and the project's configuration for every contributor. Marketplace by default, pinned snapshot as fallback.

Overview

Publisherapache
Repositorymagpie
Skill namesetup
Stars
132
Forks
120
Bundled files
57
LicenseApache-2.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 57 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by apache on GitHub. Read the source before you install it.

Installation

Install the Setup AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/apache/magpie/tree/main/plugins/magpie-setup/skills/setup
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/apache/magpie.git /tmp/magpie
mkdir -p .claude/skills
cp -r /tmp/magpie/plugins/magpie-setup/skills/setup .claude/skills/setup
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Setup in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Setup on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Setup is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

setup

The marketplace install is the default. One marketplace add, then one plugin install per skill family, and nothing is written to the repo. setup with no arguments proposes it; the other methods are fallbacks for where a marketplace cannot reach. Walk-through: install.md → Step M0b, which reads any committed floor first. Per-agent reference: docs/setup/marketplace.md.

First — the welcome guide

The setup family runs no shared pre-flight, so this skill checks for its first-run welcome guide itself, before any sub-action, silently. <skill-dir> is the directory holding this SKILL.md, which carries the checker even before config has copied it into its checker_dir. Append :<skill-dir> only when that directory is outside the project's working tree (a marketplace install's plugin cache); when this skill runs from inside the repository — the committed .agents/skills/magpie-setup/ bootstrap — leave it off, so the check never imports code the repository ships:

bash
PYTHONPATH="$HOME/.cache/apache-magpie/preflight[:<skill-dir>]" \
  python3 -P -m setup_preflight.welcome due --skill setup
  • {"verdict": "ok"}, or the command did not run at all → say nothing and go on to the sub-action.
  • {"verdict": "action", ...} → follow rules.welcome, the same rules a pre-flight welcome finding carries, and run its record command with the PYTHONPATH above.

This is the only check: none of the pre-flight's setup, floor or fingerprint findings run here.

The install paths

PathWhat it sets upTake it when
marketplace (default)The agent's own plugin mechanism installs the skills, per machine. Nothing in the repo; updates arrive through the agent's plugin update.The agent has a plugin / extension mechanism — Claude Code, Codex CLI, VS Code + Copilot, Antigravity, Cursor, microsoft/apm. The normal case.
marketplace + an adopted floorThe same per-machine install, plus a committed .apache-magpie.lock recording the project's minimum version and plugin set. Contributors are brought up to it by any skill's pre-flight.The project wants a floor every contributor meets, without pinning anyone to one version. Written by adopt.
svn-zip / git-tag / git-branch — the pinned snapshot install (fallback)The gitignored snapshot at <snapshot-dir>, both lock files, gitignored magpie-* symlinks, the overrides scaffold, the post-checkout hook.Only where a marketplace cannot reach: the agent has no plugin mechanism, the project needs the signed ASF source artefact, or it wants every contributor and CI job pinned to one committed framework version with drift detection.
localCommitted symlinks into the in-repo skills/ source. No fetch, no snapshot.The Apache Magpie framework checkout itself (see below).

A project can have contributors on either path, but one machine takes only one (Golden rule 10).

The pinned-snapshot machinery — what the snapshot is, the fetch methods, symlink naming, the overrides scaffold — is in snapshot-model.md. A marketplace install needs none of it.

The two lock files

(Pinned-snapshot path only — a marketplace install has no lock files.) The committed <committed-lock> records what the project pins to; the gitignored <local-lock> records what this machine fetched. Drift detection and multi-installer support rest on that split. Trusted external skill sources use their own pair with the same semantics. Formats, fields, and drift rules: locks.md.

Detail files in this directory

The Sub-actions table maps every invocation to the file it loads. Three more are reference files the sub-actions consult: locks.md (lock files and the reconciled: stamp), agents.md (agent-target registry, canonical-plus-relay model), and snapshot-model.md (pinned-snapshot machinery and the five golden rules that bind only there). failure-modes.md is for when something has already gone wrong.

Golden rules

Golden rule 1 — never modify the snapshot. <snapshot-dir> is a gitignored build artefact and read-only to adopters. Every change an adopter wants goes into .apache-magpie-overrides/, where it is committed and survives the next upgrade. Neither this skill nor any framework skill reading overrides ever writes to <snapshot-dir>.

Golden rule 5 — .agents/skills/ is canonical; everything else just relays into it. Whatever layout the project had before, adopt wires the framework the same way: the canonical magpie-* links live in .agents/skills/, and every other active target (.claude, .github, holdouts) gets per-skill relay symlinks back to them (.claude/skills/magpie-<n> → ../../.agents/skills/magpie-<n>). The adopter's own non-magpie- skills in those directories are left alone. See agents.md.

Golden rule 7 — agentic overrides are read at run-time — lookup order and guardrail in overrides.md.

Golden rule 8 — family membership is declared in frontmatter; two families are always installed, the rest are opt-in. Every framework skill declares its family in a family: frontmatter key (e.g. family: repo-health). The sub-actions read that key from the snapshot to build the adopt/upgrade install choice and to wire each family's symlinks. Never infer family from the skill-name prefix — families such as repo-health and contributor-growth span several prefixes. The family vocabulary is validated by skill-and-tool-validator (ALLOWED_FAMILIES) and listed for adopters in README.md → Skill families.

Two families are wired unconditionally on every adopt / upgrade / worktree-init run, without asking: setup (every family: setup skill except the bootstrap setup itself, which is copied per Rule 6) and utilities. Both are read from frontmatter like any other family — do not keep a list here.

Golden rule 10 — marketplace first; the snapshot install is a fallback, not the default. Rules 1–9 govern the pinned-snapshot machinery; this rule decides which path a run takes. Unless the user passed an explicit method:, install proposes the marketplace install, with the exact commands for the agent in front of it. It proposes the snapshot only for one of these reasons, named out loud:

  • the agent has no plugin / extension mechanism, or its marketplace install has already failed for this user;
  • the project needs the signed ASF source artefact (svn-zip), not a marketplace clone of the repo;
  • the project wants one committed pin — every contributor and CI job on one framework version, with drift detection;
  • the repo is the framework checkout itself (method:local).

Never lay the snapshot on top of a working marketplace install "to be safe". It loads a second copy of every skill, doubling the always-on token cost, and /magpie-<skill> (snapshot) and /magpie-<family>:<skill> (plugin) then resolve to two different versions of the same skill. Where a repo genuinely needs the pin, the snapshot replaces the marketplace install on that machine — uninstall the plugins first.

Five rules bind only on the pinned-snapshot path — the lock files, drift remediation, .gitignore, copy-versus-symlink, and reloading setup after a self-update (rule 9). They are in snapshot-model.md with the machinery they govern; a marketplace install is not subject to them.

Golden rule 11 — a missing vetted-ops policy never blocks a setup run. A repo that has only installed Magpie has no .apache-magpie-overrides/tools/vetted-ops/config.toml; vetted-ops then reads a maintainer's personal copy (<personal layer>/tools/vetted-ops/config.toml), and with neither a vetted-op call fails with no vetted-ops config at …. During any setup sub-action, report that as n/a in one line and carry on with the remaining steps; never call it a blocker, never stop the run for it, and never write a policy to get past it. The one sanctioned way to a personal policy is config's personal tool configs step, on the user's explicit yes. The exemption is this run's only: every other skill still stops when the dispatcher refuses, and a policy that exists but fails to load is a real error here too.

Sub-actions

The skill dispatches by the first positional argument:

InvocationLoadsPurpose
setup (no args)install.mdFirst-time install. Proposes the marketplace install first; falls back to the snapshot only per Golden rule 10, and only that fallback is main-checkout only. Idempotent — on an already-installed repo it behaves like verify.
setup installinstall.mdSame as no-arg — explicit form. Main-checkout only.
setup install method:marketplaceinstall.md → Step M0bThe default path, named explicitly. Prints the agent's marketplace add + plugin install commands and writes nothing to the repo. Works in a worktree, and in a repo that has adopted nothing.
setup install method:svn-zip|git-tag|git-branchinstall.mdThe pinned snapshot install — the fallback path. Main-checkout only.
setup configconfig.mdNot an install, and not adoption. Configure the installed skills for you, in the personal layer — <git-common-dir>/apache-magpie/ on a repo that has not adopted Magpie (nothing in the working tree), the gitignored .apache-magpie-local/ on one that has. Works on any repo, adopted or not, with nobody's permission. Writes nothing committable and stages nothing.
setup config <skill>config.mdThe same, narrowed to one skill's required configuration.
setup config adversarial-reviewconfig.mdDetect the installed model CLIs and configure them as adversarial reviewers.
setup adoptadopt.mdNot an install. Commit the repo's recommended default plugin set and scaffold its overrides store, so every contributor arrives with them. Needs an explicit maintainer decision; stages, never commits. Default set is Claude Code only.
setup upgradeupgrade.mdRefresh the snapshot per <committed-lock>, reconcile overrides, refresh symlinks. Main-checkout only — worktrees pick it up through the symlink worktree-init installs.
setup worktree-initworktree-init.mdWorktree-only. Symlink the worktree's <snapshot-dir> to the main checkout's so both share one framework state. No fetch, no lock files written; idempotent.
setup verifyverify.mdRead-only health check and drift report. Works in the main checkout and in worktrees.
setup reconcilereconcile.mdOne-time project-wide sweep: checks anchors and requires_config for every configured or overridden skill, proposes fixes one by one, writes the reconciled: stamp. Any install method. Main-checkout only when the stamp goes to the committed lock (adopted); unrestricted when it goes to the local file.
setup skill-sources (aka skill-sources add <id>)skill-sources.mdFetch, verify, pin and symlink skills from the trusted sources listed in <project-config>/skill-sources.md. Main-checkout only — worktrees share the source snapshots via worktree-init.
setup override <skill>overrides.mdOpen or scaffold an override file.
setup delegate [<skill>]delegates.mdFind the project's own skills that overlap Magpie's and, per match and on the user's yes, write a delegate: block so Magpie runs the project's skill inside its workflow. Magpie's wrapper — confirmation gates, scrubs, the privacy gate — stays around it.
setup welcome [family]welcome.mdAlias for the welcome-guide skill: show a welcome guide, or turn one back on after a Dismiss or Don't show again.
setup uninstalluninstall.mdReverse the install: remove the snapshot, the local lock, symlinks, hook, doc sections, and this skill. Leaves .apache-magpie.lock — that is unadopt. Keeps .apache-magpie-overrides/ unless --purge-overrides is passed. Main-checkout only.
setup unadoptadopt.md → UnadoptRemove the committed floor lock .apache-magpie.lock and the .claude/settings.json wiring derived from it. Keeps .apache-magpie-overrides/ unless --purge-overrides is passed. Leaves every install — yours and everyone else's — untouched.

Main-checkout-only is marked per row above. Those sub-actions compare git rev-parse --git-dir with --git-common-dir and, in a worktree, refuse and point back at the main checkout. The marker has two reasons: upgrade and the snapshot fallback of install/uninstall are pinned-snapshot operations, while adopt/unadopt write committed files. A marketplace install touches no repo state and runs anywhere.

reconcile is restricted only when the project is adopted, because its stamp then goes to the committed lock. A configured-but-unadopted project writes reconciled.json in its personal layer, <git-common-dir>/apache-magpie/, and has no worktree restriction (reconcile.md Step 0).

adopt and upgrade always chain into worktree-init on every linked worktree. It is a no-op with no worktrees and idempotent where they already look wired, which is how broken symlinks and newly always-on families get repaired — nobody has to re-run anything per worktree.

A missing snapshot with a committed lock turns any sub-action into the recover-snapshot path: re-install per the lock, then continue.

Unrecognised sub-action

When the first positional argument matches no sub-action in the table above and is not a flag from Inputs, do not guess, and do not fall back to install. The usual cause is a typo (upgrede, reconcie), and quietly mapping it to the nearest name runs something the user never named — upgrade, reconcile and adopt stage committed files.

Print this instead, using the invocation name this install answers to (/magpie-setup:setup on a marketplace install, /magpie-setup on the pinned snapshot):

text
Unknown setup sub-action: `upgrede`.
Did you mean `upgrade`?  →  /magpie-setup:setup upgrade

Sub-actions: install, config, adopt, upgrade, worktree-init, verify,
reconcile, skill-sources, override, delegate, welcome, uninstall, unadopt
  • Suggest only a close match: an edit distance of one or two, or a prefix of the name. When several qualify (un → uninstall, unadopt), name them all. When none is close, print only the list.
  • Then stop and wait. Run a suggested sub-action only once the user confirms it — a plain "yes" to a single suggestion is enough — never on your own because it was the only near match.

Inputs

FlagEffect
from:<git-ref> / from:<version>Install or upgrade from a specific framework ref or version. On install it replaces the prompt; with method:marketplace it pins the marketplace to that tag (/plugin marketplace add apache/magpie@<version>). On upgrade it overrides the committed lock for this run only and does NOT update it.
method:<marketplace|git-branch|git-tag|svn-zip|local>Pick the install method explicitly. Default on install: marketplace. The others are fallbacks (Golden rule 10), so the agent proposes the marketplace path and names the fallback instead of opening with a three-way prompt. marketplace writes nothing to the repo (install.md → Step M0b). local is framework-checkout only: it self-adopts by linking the in-repo skills/ source instead of fetching a snapshot (install.md → Local self-adoption).
agents:<list>Comma-separated agent targets to wire symlinks into (agents.md registry ids: universal, claude-code, github, windsurf, goose, …). Default on adopt/upgrade: the always-on neutral set (universal + claude-code + github) plus any other registry dir already in the repo. When passed, it replaces the detected set for that run — except universal (.agents/skills/), which is always kept because every other target relays into it.
skill-families:<list>Comma-separated opt-in families: what to symlink on the snapshot path, which magpie-<family> plugins to install on the marketplace path. Any opt-in family declared by a family: key in the snapshot (today: security, pr-management, issue, release-management, repo-health, pairing, mentoring, contributor-growth). Default on adopt: prompt (install.md Step 5). Default on upgrade: the families in <committed-lock> / <local-lock>, plus any opt-in family added since the lock was written (recorded back into the lock), and every framework skill in that set gets a valid symlink — missing or broken ones are created or repaired. Never accepts the always-on families (setup, utilities); per Golden rule 8 they are wired on every run and cannot be requested or opted out of.
--purge-overrides(unadopt and uninstall) Also git rm -r .apache-magpie-overrides/. Default on both: keep.
--no-overrides(any framework skill) Skip override lookup for this one invocation and run on framework defaults. Override files on disk are not read, changed, or deleted. The safety baseline (confidentiality, privacy, security) still applies. See One-shot defaults run.
dry-runShow what the skill would do without writing anything.

What this skill is NOT for

Failure modes

What each sub-action does when the ground is not what it expected, and which cases are recoverable: failure-modes.md.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Setup AI skill do?

Install Magpie, configure it for yourself, or adopt it for a repo. Installing touches only this machine; configuring writes gitignored local files; adopting commits a floor and the project's configuration for every contributor. Marketplace by default, pinned snapshot as fallback.

Why use Setup on TypingMind?

Because you install it once and use it with any model. Setup is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Setup in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/apache/magpie/tree/main/plugins/magpie-setup/skills/setup. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Setup?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Setup?

As many as you like. As long as a model supports skills, you can use Setup with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Setup AI skill free?

Yes. It is published on GitHub by apache under the Apache-2.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇