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. GitHub Copilot helps you write code in the editor, and its Code Review feature reviews PRs. CodeRabbit and PR-Agent (Qodo) automate PR workflows with line-level suggestions and descriptions. But what about everything else? Implementing issues, bumping dependencies, diagnosing CI, monitoring your community, and iterating on fixes?

Copilot codes with you. CodeRabbit and PR-Agent review PRs. RepoKeeper runs your repo while you sleep.

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 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• 🔍 Daily Patrol — Scans 8 ecosystems (pip, npm, Go, Cargo, Bundler, Composer, Maven, Gradle) for outdated deps.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. Supports two backends: native (single LLM call, ~$0.001) and Pi (autonomous agent loop, reads files, runs tests, self-corrects). PR fix mode: comment /repokeeper go on a PR with feedback → agent reads the conversation → pushes fixes to the same branch.
• 📝 Code Review Agent — Reads PR diffs → checks against your profile → posts inline line-level comments with severity indicators and suggestion blocks. Incremental re-review on new commits. Auto-generates PR descriptions from diffs.
• 🏷️ 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. Diff-aware PR classification.
• 👤 Maintainer Profile — One YAML file. Code style, tone, PR standards, tech stack preferences, skip keywords, verification commands. Or skip it — defaults work.

Architecture

flowchart TD
    subgraph T["Triggers"]
        direction LR
        issues["🐙 Issues"]
        prs["🐙 Pull Requests"]
        disc["🐙 Discussions"]
        cmd["💬 /repokeeper"]
    end

    T --> M

    subgraph M["RepoKeeper Modules"]
        direction LR
        radar["🔭 Radar<br/>community monitoring"]
        patrol["🔍 Patrol<br/>deps · CI · stale"]
        agent["🤖 Agent<br/>native · Pi backend"]
        labeler["🏷️ Labeler<br/>auto-classify"]
        review["📝 Review<br/>inline comments"]
    end

    profile["📋 repokeeper.yml"]:::config
    profile -.-> M

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

    subgraph O["Outputs"]
        direction LR
        pr["✅ PRs"]
        rev["📝 Reviews"]
        lbl["🏷 Labels"]
        notify["📬 Notify"]
    end

    M --> O

    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]
  pull_request_target:
    types: [labeled]

jobs:
  agent:
    runs-on: ubuntu-latest
    if: |
      (
        github.event_name == 'issue_comment' &&
        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'
      ) ||
      (
        github.event_name == 'pull_request_target' &&
        github.event.label.name == 'agent-fix'
      )
    permissions:
      contents: write
      issues: write
      pull-requests: write
    steps:
      - name: Determine context
        id: ctx
        shell: bash
        run: |
          if [ "${{ github.event_name }}" = "pull_request_target" ]; then
            echo "issue=${{ github.event.pull_request.number }}" >> "$GITHUB_OUTPUT"
            echo "pr=${{ github.event.pull_request.number }}" >> "$GITHUB_OUTPUT"
          elif [ "${{ github.event.issue.pull_request }}" != "" ]; then
            echo "issue=${{ github.event.issue.number }}" >> "$GITHUB_OUTPUT"
            echo "pr=${{ github.event.issue.number }}" >> "$GITHUB_OUTPUT"
          else
            echo "issue=${{ github.event.issue.number }}" >> "$GITHUB_OUTPUT"
            echo "pr=" >> "$GITHUB_OUTPUT"
          fi

      - uses: shenxianpeng/repokeeper/agent@v1
        with:
          repo: ${{ github.repository }}
          issue: ${{ steps.ctx.outputs.issue }}
          pr: ${{ steps.ctx.outputs.pr }}
          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.
• Label a PR agent-fix — or comment /repokeeper go with feedback on a PR.
• Comment /repokeeper review on a PR for inline code review.

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

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 and inline code review comments; it does not approve or merge 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