【Agent工坊】AI机器人正在淹没开源仓库：Archestra团队血泪防守实录与可复用工具链

# London-Cat 核心逻辑简化版

import requests

import os

GITHUB_TOKEN = os.environ["GITHUB_TOKEN"]

REPO = "archestra-ai/archestra"

def get_contributor_reputation(username: str) -> dict:

    """计算贡献者在目标仓库的信誉分"""

    headers = {"Authorization": f"Bearer {GITHUB_TOKEN}"}

    # 1. 统计 merged PRs

    merged_prs = requests.get(

        f"https://api.github.com/search/issues"

        f"?q=repo:{REPO}+is:pr+author:{username}+is:merged",

        headers=headers

    ).json().get("total_count", 0)

    # 2. 检查是否被标记为 spam

    spam_labels = requests.get(

        f"https://api.github.com/search/issues"

        f"?q=repo:{REPO}+author:{username}+label:spam",

        headers=headers

    ).json().get("total_count", 0)

    # 3. 计算综合分数

    score = min(100, merged_prs * 20 - spam_labels * 50)

    return {

        "username": username,

        "merged_prs": merged_prs,

        "spam_count": spam_labels,

        "reputation_score": max(0, score),

        "trust_level": "high" if score >= 60 else "low" if score < 20 else "medium"

    }

# 在 issue comment webhook 中调用

def on_issue_comment(comment):

    rep = get_contributor_reputation(comment["user"]["login"])

    if rep["trust_level"] == "low" and rep["merged_prs"] == 0:

        # 自动标记为可疑

        label_comment_as_spam(comment["id"])

        post_warning(comment["issue_url"],

                     f"⚠️ @{comment['user']['login']} 信誉分 {rep['reputation_score']}，"

                     f"此评论已被标记为待审核。")

# GitHub Actions workflow: AI Sheriff

name: AI Spam Detector

on:

  issues:

    types: [opened, edited]

  pull_request:

    types: [opened, edited]

  issue_comment:

    types: [created, edited]

jobs:

  detect-spam:

    runs-on: ubuntu-latest

    steps:

      - uses: actions/checkout@v4

      - name: Analyze content with AI

        id: analyze

        uses: archestra-ai/spam-detector-action@v1

        with:

          openai-key: ${{ secrets.OPENAI_API_KEY }}

          content: ${{ github.event.comment.body || github.event.issue.body }}

          thresholds: |

            spam_probability: 0.85

            ai_generated_probability: 0.90

            min_account_age_days: 7

      - name: Auto-close if spam

        if: steps.analyze.outputs.is_spam == 'true'

        uses: actions/github-script@v7

        with:

          script: |

            const reason = `${{ steps.analyze.outputs.reason }}`;

            const spamLabel = '🤖 AI-generated';

            if (context.eventName === 'pull_request') {

              await github.rest.issues.createComment({

                ...context.repo,

                issue_number: context.issue.number,

                body: `⚠️ 此 PR 被 AI Sheriff 检测为疑似 AI 生成内容。\n原因：${reason}\n\n如果你是真实开发者，请回复 "!human" 触发人工审核。`

              });

              await github.rest.pulls.update({

                ...context.repo,

                pull_number: context.issue.number,

                state: 'closed'

              });

            }

            await github.rest.issues.addLabels({

              ...context.repo,

              issue_number: context.issue.number,

              labels: [spamLabel]

            });

# onboarding.py — Archestra 贡献者白名单系统的简化实现

import hashlib

import hmac

import json

import os

import time

import requests

from flask import Flask, request, jsonify

app = Flask(__name__)

GITHUB_TOKEN = os.environ["GITHUB_TOKEN"]

WEBHOOK_SECRET = os.environ["WEBHOOK_SECRET"]

REPO = "archestra-ai/archestra"

# 使用 GitHub 的 co-author 机制将用户标记为"prior contributor"

@app.route("/onboard", methods=["POST"])

def onboard_contributor():

    """五步入门流程，完成后自动创建一个 co-authored commit"""

    data = request.json

    github_username = data.get("github_username")

    if not github_username:

        return jsonify({"error": "Missing github_username"}), 400

    # 步骤1-4：CAPTCHA + 行为准则确认 + 文档阅读 + 小测验

    # （此处省略前端流程，假设已通过）

    # 步骤5：创建 co-authored commit 将用户标记为 prior contributor

    result = create_onboarding_commit(github_username)

    return jsonify({

        "status": "onboarded",

        "commit_sha": result.get("sha"),

        "message": "你现在可以创建 issues、PR 和评论了！"

    })

def create_onboarding_commit(username: str) -> dict:

    """利用 GitHub co-author 机制创建白名单 commit"""

    # 1. 获取当前 CONTRIBUTORS.md 的 SHA

    headers = {

        "Authorization": f"Bearer {GITHUB_TOKEN}",

        "Accept": "application/vnd.github+json"

    }

    file_resp = requests.get(

        f"https://api.github.com/repos/{REPO}/contents/CONTRIBUTORS.md",

        headers=headers

    ).json()

    # 2. 在 CONTRIBUTORS.md 末尾追加用户名（实际操作用无害的微小改动）

    current_content = requests.get(file_resp["download_url"]).text

    new_content = current_content + f"\n- @{username} (onboarded {time.strftime('%Y-%m-%d')})"

    # 3. 创建 commit，使用 co-author 将用户标记为 contributor

    commit_payload = {

        "message": f"docs: onboard @{username} as contributor\n\n"

                   f"Co-authored-by: {username} <{username}@users.noreply.github.com>",

        "content": base64_encode(new_content),

        "sha": file_resp["sha"],

        "branch": "main",

        "committer": {

            "name": "Onboarding Bot",

            "email": "bot@archestra.ai"

        },

        "author": {

            "name": username,

            "email": f"{username}@users.noreply.github.com"

        }

    }

    resp = requests.put(

        f"https://api.github.com/repos/{REPO}/contents/CONTRIBUTORS.md",

        headers=headers,

        json=commit_payload

    )

    return resp.json()

def base64_encode(content: str) -> str:

    import base64

    return base64.b64encode(content.encode()).decode()

if __name__ == "__main__":

    app.run(port=5000)

#!/bin/bash

# .github/scripts/check-first-timer.sh

# PR 创建时运行，检查作者是否在仓库历史中

AUTHOR_EMAIL="$1"

MIN_COMMITS=2

EXEMPT_LABEL="legitimate-newcomer"

# 例外：手动标记的合法新人

if gh pr view --json labels | grep -q "$EXEMPT_LABEL"; then

    echo "✅ 已标记为合法新人，跳过检查"

    exit 0

fi

# 检查该 email 在仓库历史中的 commit 数量

COMMIT_COUNT=$(git log --all --author="$AUTHOR_EMAIL" --oneline | wc -l)

if [ "$COMMIT_COUNT" -lt "$MIN_COMMITS" ]; then

    echo "⚠️ 警告：$AUTHOR_EMAIL 在仓库中仅有 $COMMIT_COUNT 个 commits"

    echo " 此 PR 被自动标记为待审核"

    gh pr edit --add-label "needs-review"

    gh pr comment --body "🤖 自动检测：你在本仓库的 commit 数量不足 $MIN_COMMITS。\n

维护者将在 24 小时内审核。为加快流程，请在 PR 描述中说明你的实现思路。"

else

    echo "✅ $AUTHOR_EMAIL 有 $COMMIT_COUNT 个历史 commits，通过检查"

fi

# 1. 开启 GitHub 仓库的互动限制

# Settings → Moderation → Interaction limits →

# "Limit to prior contributors" (临时模式)

# 2. 在 CI 中添加作者历史检查

# 将上面的 check-first-timer.sh 加入 PR workflow

# 3. 设置 issue template 要求贡献者描述实现思路

# .github/ISSUE_TEMPLATE/feature_request.md 中添加：

# "请用至少50字描述你的实现思路（AI生成的泛泛而谈将被关闭）"