MStack: Open Source Maintainer Automation for Claude Code

Open source maintainer automation skills for Claude Code.

An OSS maintainer with 50 unread issues spends 80% of their time on grunt work: labeling, finding duplicates, screening spam PRs, drafting responses. MStack compresses that to near-zero. The judgment stays with the maintainer.

Skills

Skill	What it does	Say this to Claude
maintain	Full pipeline: triage + review + respond + health	"run mstack maintain"
triage	Categorize issues, detect duplicates, flag bots	"triage the issues in this repo"
review-prs	Pre-screen PRs: security, tests, quality, CI	"review the open PRs"
respond	Draft tone-matched responses for issues and PRs	"draft responses for open issues"
health	Repo health report: stale, CI, dependencies	"run a health check on this repo"
mstack-release	Changelog + version bump + GitHub release	"cut a new release"
mstack-setup	Configure MStack for this repo	"setup mstack for this repo"

Install (30 seconds)

git clone --single-branch --depth 1 https://github.com/Abhilash-003/mstack.git
cd mstack && ./setup

The setup script symlinks all skills into ~/.claude/skills/ so Claude Code can discover and invoke them. Works regardless of where you clone the repo.

Quick Start

MStack skills are invoked through natural language in Claude Code. Just describe what you want:

"run the full maintenance pipeline on this repo"
"triage the open issues"
"review open pull requests"  
"draft responses for unanswered issues"
"give me a health report for this repo"
"cut a new release"

You can also be specific:

"review PR #42"
"respond to issues #17 and #23"

First time? Start with:

"setup mstack for this repo"

The Maintenance Pipeline

TRIAGE → REVIEW → RESPOND → HEALTH
  ↑          ↑        ↑        ↑
  └──────────┴────────┴────────┘
         human checkpoints

Every phase transition is a checkpoint. You approve the triage actions before any labels are applied. You review each PR analysis before a comment is posted. You read each draft response before it goes live. The pipeline is assistive, not autonomous.

How It Works

Each skill is a SKILL.md file that Claude Code reads and follows. No backend, no database, no custom agents. Logs and reports live in .mstack/ at your project root. Global config lives in ~/.mstack/.

All GitHub actions happen through the gh CLI — which you're already authenticated with. No Personal Access Token, no OAuth app, no secret management.

Architecture

• Pure SKILL.md files — no Express, no React, no Postgres. Claude Code IS the runtime.
• Logs at project root — structured JSONL in .mstack/logs/, health reports in .mstack/reports/.
• gh CLI for everything — already authenticated, no extra credentials needed.
• Two-phase install — offline bootstrap (./setup) + interactive per-repo setup (say "setup mstack").
• Human-in-the-loop — every label, comment, close, tag, and push requires explicit approval.

See ARCHITECTURE.md for the full design rationale.

Requirements

• Claude Code (or any Claude Code-compatible agent)
• gh CLI (brew install gh or cli.github.com) — must be authenticated
• git

Project State

Internal plumbing lives in .mstack/ at your project root.

your-repo/
└── .mstack/
    ├── config.yml                      # Repo identity and behavior config
    ├── logs/
    │   ├── triage-2026-04-13.jsonl     # Issue analysis (append-only)
    │   ├── review-2026-04-13.jsonl     # PR analysis (append-only)
    │   └── maintain-2026-04-13.json    # Session state
    └── reports/
        └── 2026-04-13-health.md        # Health check report

Add .mstack/ to .gitignore to keep logs local (MStack will offer to do this during setup).

Configuration

Global config at ~/.mstack/config.yaml:

bin/mstack-config get stale_days         # read: 30
bin/mstack-config set stale_days 60      # write
bin/mstack-config list                   # show all

Per-repo config at .mstack/config.yml — written by mstack-setup, editable directly.

Key settings:

Key	Default	Effect
detect_duplicates	true	Semantic duplicate detection during triage
detect_bots	true	Bot/spam flagging in triage and PR review
stale_days	30	Days before an issue or PR is considered stale
require_tests	true	Flag PRs that change source without adding tests
security_scan	true	Scan PR diffs for hardcoded secrets and injection risks
max_files_warn	50	Warn when a PR touches more than N files
changelog_format	keep-a-changelog	Changelog style for release skill
version_scheme	semver	Version bump strategy

Comparison

	MStack	GitHub Agentic Workflows	PR-Agent	Manual
Scope	Full maintenance pipeline	CI automation only	PR review only	Everything
Infrastructure	None (SKILL.md files)	GitHub Actions runners	Hosted service or self-host	None
Human approval	Every action	Configurable, often none	Optional	Always
Issue triage	Yes	No	No	Manual
Response drafting	Yes (tone-matched)	No	No	Manual
Repo health reports	Yes	No	No	Manual
Release automation	Yes	Partial	No	Manual
Auth required	gh CLI (existing)	GitHub Actions token	API key	None
Install	30 seconds	Per-workflow	Complex	N/A

Documentation

• README.md — this file
• ARCHITECTURE.md — why MStack is built this way
• CLAUDE.md — development commands, project structure, config reference
• CONTRIBUTING.md — how to add skills and contribute
• CHANGELOG.md — release notes

License

MIT