Apache Magpie™ Documentation
Apache Magpie is an AI assistant for open-source project maintainers. It handles the repetitive parts of running a project — triaging issues, reviewing PRs, onboarding contributors, managing security reports, cutting releases — so maintainers can spend their time on the work that needs a human. The agent proposes; the human decides.
⚡ Quick Start
Install Apache Magpie into the agent you already use — one command to add the Apache Magpie Marketplace, one per family you want. That install is yours, on this machine: nothing is committed to your repository, and it needs no decision from your project and no opt-in from your teammates.
Adoption is the other thing, and it belongs to the project. It is not an install but a maintainer decision — the team agreeing which families they recommend here and how those should behave in this project, then committing that floor so everyone who clones the repo picks it up on arrival. It affects every contributor, so it is the project's call to make together — and you do not have to choose before installing.
- Installation or Adoption?
- What each family solves
- Your first run with a family
- Other installation methods
- Prerequisites for running framework skills
Not sure what Magpie is yet? What is Apache Magpie? · Architecture
Explore the docs
154 pages, synced from apache/magpie/docs and grouped by area. Every page is also in the sidebar.
Concepts
What Magpie is and the ideas it is built on — the design principles, the agentic modes and what they cost, labels and capabilities, vendor neutrality, and confidentiality.
Updating Magpie
Working on the framework itself — the spec-driven build loop, the editorial playbook for text it produces, how contributor sentiment is measured, the pilot report template, what ships in a source release, and how to extend Magpie without forking it.
Education
The maintainer-education stream — a step-by-step path to building and running agents for your project: what agents are, working with them, writing and evaluating your first skill, English as code, and contributing back, plus a pattern catalogue and hands-on tutorials.
Training
An LMS-neutral Apache Training module that wraps the education stream as a structured course — one lesson per progression page, each adding learning objectives, exercises, and self-check questions.
- lesson 01 what agents are
- lesson 02 working with agents
- lesson 03 choosing models
- lesson 04 your first skill
- lesson 05 writing safe skills
- lesson 06 debugging a skill
- lesson 07 writing portable skills
- lesson 08 eval driven development
- lesson 09 agentic and autonomous work
- lesson 10 english as a programming language
- lesson 11 how to contribute
- lesson lab tutorials
- instructor guide
- upstream contribution
Setup
Wire Magpie into a repo and run agents safely — the marketplace, individual use versus team adoption, sandboxing, per-role MCP access, privacy/LLM routing, and uninstalling.
PR management
Maintainer-facing PR-queue triage, stats, and deep code review.
Issue management
Issue triage, bug reproduction, fix drafting, and backlog re-assessment.
Mentoring
Contributor mentoring — the spec, tone guide, and first skills.
Contributor growth
The contributor-to-committer path — welcoming, activity tracking, and nomination evidence.
Repo health
Read-only maintenance audits — CI, dependencies, licenses, and flaky tests.
Pairing
Developer-side skills you run in your own dev loop.
Release management
The ASF release lifecycle — plan, cut, vote, promote, announce.
Security
The security-issue lifecycle — process, roles, forwarder routing policy, onboarding, and the threat model. Maintainer-only.
Utilities
Framework meta-skills — author skills and list every skill.
Adapters
Authoring adapters and the discovery registry of in-tree and external backends.
Tools
Every framework tool — the contract or substrate it provides, with per-tool docs.
- agent-guard
- agent-isolation
- apache-projects
- asf-svn
- bitbucket
- change-request
- cve-org
- cve-tool
- cve-tool-vulnogram
- dashboard-generator
- dev
- egress-gateway
- forwarder-relay
- fossil
- github
- github-body-field
- github-rollup
- gmail
- jira
- jira-patch
- mail-archive
- mail-patch
- mail-source
- maildir
- permission-audit
- pilot-report-validator
- ponymail
- pr-management-stats
- preflight-audit
- privacy-llm
- probe-templates
- sandbox-lint
- scan-format
- security-tracker-stats-dashboard
- skill-and-tool-validator
- skill-evals
- skill-reconciler-diff
- skill-token-count
- sourcehut
- spec-inventory
- spec-loop
- spec-status-index
- spec-validator
- symlink-lint
- vcs
- vendor-neutrality-score
- vetted-ops
RFCs
Design proposals under discussion.
- RFC-AI-0002: Secure Agents setup
- RFC-AI-0003: Privacy-aware LLM routing for foundation private information
- RFC-AI-0004: Principles of agentic interaction for open-source maintainers
- RFC-AI-0005: Framework taxonomy
- RFC-AI-0006: Trusted external skill sources
- RFC-AI-0007: Selective snapshot — fetch only the chosen skill families
- RFC-AI-0008: The coordinator role — multi-hop relay of security reports
Designs
External Skill Sources
Pulling Magpie-shaped skills from a repository other than apache/magpie — the trust model, the registry of vetted sources, and how to author one.