repokeeper 1.2.0

AI-powered open source maintainer agent

RepoKeeper

AI-powered open source maintainer agent. Reads issues, writes code, opens PRs — 24/7.

# Label an issue agent-todo — RepoKeeper handles the rest
/repokeeper go

Zero config. GitHub-native. ~$0.01 per PR with DeepSeek.

---

Why RepoKeeper?

Open source maintenance is a second job you didn't sign up for. AI coding agents like Copilot and Cursor help you write code in the editor. PR Agent (Qodo) automates PR workflows. But what about everything else? Triaging issues, bumping dependencies, diagnosing CI, responding to the community?

Copilot codes with you. PR Agent polishes PRs. RepoKeeper runs your repo while you sleep.

	Copilot / Cursor	PR Agent (Qodo)	RepoKeeper
What it does	AI agent coding in your editor	Automates PR descriptions, reviews, suggestions	Maintains your repo autonomously
How	Reads codebase → implements → verifies	PR lifecycle automation	Reads issues + codebase → opens verified PRs
When	While you code	On PR events	24/7 on schedule (labels, comments, cron)
Community	No	No	Monitors, classifies, responds
Dependencies	No	No	Scans 8 ecosystems for outdated deps
CI	No	No	Diagnoses failures, suggests fixes
Cost	$10–39/month subscription	Free OSS / paid plans	~$0.01 per PR (your own LLM key)
Config	IDE settings	CLI / PR comments	One YAML (or zero)

What It Does

• 🔭 Community Radar — Monitors GitHub issues and discussions for keywords. AI classifies hits as bugs, feature requests, or noise. Auto-creates issues with deduplication and RepoKeeper branding, linking back to original discussions. Notifies you via email, Telegram, or WeChat.
• 🔍 Daily Patrol — Scans 8 ecosystems (pip, npm, Go, Cargo, Bundler, Composer, Maven, Gradle) for outdated deps. Diagnoses CI failures with real job/step data. Auto-fixes CI by opening repair PRs. Finds stale issues. Health score every weekday morning.
• 🤖 Implementation Agent — Reads your codebase + issue → implements → verifies (lint + tests) → pushes branch → opens PR. Streams LLM output in real-time. Estimates token cost. Supports DeepSeek, OpenAI, and Anthropic Claude models.
• 🏷️ Auto-Labeler — AI classifies new issues and PRs, picks labels from your repo's existing set (matching naming conventions), and creates new labels only when needed — with consistent style and descriptions. Supports issue and PR labeling with diff-aware classification.
• 👤 Maintainer Profile — One YAML file describing your code style, tone, PR standards. Or skip it — defaults work.

Architecture

flowchart TD
    issues["🐙 Issues"]
    prs["🐙 Pull Requests"]
    disc["🐙 Discussions"]
    comments["💬 /repokeeper"]

    profile["📋 Maintainer Profile<br/>repokeeper.yml"]:::config

    radar["🔭 Community Radar<br/>monitors community"]
    patrol["🔍 Daily Patrol<br/>deps · CI · stale issues"]
    agent["🤖 Implementation Agent<br/>issue → implement → PR"]
    labeler["🏷️ Auto-Labeler<br/>classify issues & PRs"]
    reviewer["📝 Code Review Agent<br/>inline PR review"]

    llm["🧠 LLM<br/>DeepSeek · OpenAI · Claude"]

    newpr["✅ Verified PRs"]
    review["📝 Inline Reviews"]
    applied["🏷 Applied Labels"]
    notify["📬 Email · Telegram · WeChat"]

    issues --> radar & agent & labeler
    prs --> reviewer & labeler
    disc --> radar
    comments --> agent & reviewer

    profile -. "configures" .-> radar
    profile -.-> patrol
    profile -.-> agent
    profile -.-> labeler
    profile -.-> reviewer

    radar & patrol & agent & labeler & reviewer --> llm
    llm --> radar & patrol & agent & labeler & reviewer

    agent & patrol --> newpr
    reviewer --> review
    labeler --> applied
    radar & patrol --> notify

    classDef config fill:#f9f0ff,stroke:#6e40c9,stroke-width:2px

Adopt in 60 Seconds

Three ways to onboard — pick one:

📋 Copy a workflow

Create .github/workflows/repokeeper.yml in your repo:

name: RepoKeeper Implementation Agent

on:
  issue_comment:
    types: [created]
  issues:
    types: [labeled]

jobs:
  agent:
    runs-on: ubuntu-latest
    if: |
      (
        github.event_name == 'issue_comment' &&
        !github.event.issue.pull_request &&
        contains(github.event.comment.body, '/repokeeper go') &&
        (
          github.event.comment.author_association == 'OWNER' ||
          github.event.comment.author_association == 'MEMBER' ||
          github.event.comment.author_association == 'COLLABORATOR'
        )
      ) ||
      (
        github.event_name == 'issues' &&
        github.event.label.name == 'agent-todo'
      )
    permissions:
      contents: write
      issues: write
      pull-requests: write
    steps:
      - uses: shenxianpeng/repokeeper/agent@v1
        with:
          repo: ${{ github.repository }}
          issue: ${{ github.event.issue.number }}
          llm_api_key: ${{ secrets.DEEPSEEK_API_KEY }}
          llm_base_url: ${{ secrets.LLM_BASE_URL || 'https://api.deepseek.com' }}
          github_token: ${{ secrets.REPOKEEPER_GITHUB_TOKEN || secrets.GITHUB_TOKEN }}

Then add your API key: Settings → Secrets → Actions → New secret: DEEPSEEK_API_KEY = sk-...

Before pushing, run the setup check:

repokeeper doctor --repo owner/repo

doctor verifies the profile, workflow triggers, workflow permissions, token environment, LLM key, and repository slug. Fix anything marked missing, then push the workflow.

Want Radar, Patrol, Labeler, and Review too? Add radar, patrol, labeler, and review composite actions to separate workflow files in .github/workflows/.

🖥️ CLI

pip install repokeeper
repokeeper init --all-workflows   # profile + all 5 workflows
repokeeper init --minimal         # profile + agent workflow only
repokeeper doctor --repo owner/repo

🤖 Ask AI

Paste this into any AI coding agent (Copilot Chat, Claude Code, Cursor, Windsurf, pi, etc.):

Add RepoKeeper to this repository. Create .github/workflows/repokeeper.yml that uses the shenxianpeng/repokeeper/agent@v1 composite action — trigger on issue comments (/repokeeper go) and labels (agent-todo). Pass DEEPSEEK_API_KEY as the llm_api_key input. Then tell me to add a DEEPSEEK_API_KEY secret in GitHub Actions settings.

Trigger the agent

Label any issue agent-todo — or comment /repokeeper go.

---

Install (optional CLI)

pip install repokeeper

repokeeper init             # Create a profile
repokeeper init --minimal   # Create a profile + agent workflow
repokeeper doctor --repo owner/repo
repokeeper radar --repo owner/repo
repokeeper patrol --repo owner/repo --summary
repokeeper agent --repo owner/repo --issue 42
repokeeper labeler --repo owner/repo --issue 42
repokeeper labeler --repo owner/repo --pr 42
repokeeper review --repo owner/repo --pr 42
repokeeper describe --repo owner/repo --pr 42

---

Documentation

Full docs at shenxianpeng.github.io/repokeeper

Guide
Quick Start	5-minute setup
Security	Permissions, tokens, and automation boundaries
Dogfood Cases	Public proof template for real RepoKeeper-generated PRs
Community Radar	Monitor your community
Daily Patrol	Automated health checks
Implementation Agent	AI-powered PRs
Auto-Labeler	AI-powered issue & PR labeling
Maintainer Profile	Full config reference
Code Review Agent	Inline review, PR descriptions, incremental re-review
Benchmarks	Cost and performance estimates by scenario

Contributing

Contributions are welcome, especially documentation examples, setup diagnostics, tests, and safety improvements. See CONTRIBUTING.md before opening a pull request.

Safety Model

RepoKeeper creates reviewable pull requests; it does not merge them for you. The default workflow limits write access to branches, issue comments, and pull requests, and the agent blocks edits under .github/workflows/. See the Security guide before enabling it on sensitive repositories.

---

License

MIT © Xianpeng Shen