Files
MediaCrawler/mac-agent-os-main/archive_docs/matrix_docs/AGENT_INIT_GUIDE.md
T
butubb 2112c1a870
Deploy VitePress site to Pages / build (push) Canceled after 0s
Deploy VitePress site to Pages / Deploy (push) Canceled after 0s
feat(monitor): 抖音 Web 接口客户端 —— 绕开爬虫子进程,直接发 HTTP
移植自 mac-agent-os 的 mediacrawler_adapter:不起子进程、不开页面,用浏览器里那份
登录态直接调抖音 Web 接口。产物键名照抄 store/douyin,所以 ingest 那条链路一个字不用改。

**目前能用的(真环境实测,非推断)**:

    profile/other : 200, 7075 字节  —— 博主主页指标(粉丝/获赞/作品数/昵称)
    aweme/detail  : 200, 45425 字节 —— 单条作品详情(含点赞/评论/收藏/分享)

**目前不能用的:作品列表 `aweme/post`。** 两个互相独立的原因:
  1. 这个接口被抖音单独升级成了真校验:不带 x-tt-argus 回 403「Uifid Not Found」,
     带上 dummy 值回 200 + **空 body**。也就是说「头在不在」骗得过,「真校验」过不了。
     同一套头打 profile/other 和 aweme/detail 都是通的 —— 抖音是挑着接口加保护的,
     挑中的恰好是「批量拉作品列表」这个最敏感的动作。
  2. 改走页面截获也不行:CDP 浏览器打开博主主页会落到「验证码中间页」(当天大量探测的
     代价,过几小时要重测)。

  所以现在的边界是:**已知作品的指标刷新能做,自动发现新作品做不了**。

**排查中控住变量后得到的两条事实**(都写进注释了):
  · `Accept` / `Accept-Language` / `Referer` 才是主页接口能返回真数据的原因 —— 只有
    UA+client hints+Cookie 时是 200 但仅 121 字节的空壳,补上这三个头变 7074 字节。
    (我先前猜的 sec-ch-ua 不是关键。)
  · 因此 UA 与 client hints 必须**成套地取自同一个浏览器**,所以 BrowserIdentity 一次
    从 CDP 取齐 cookie + UA + hints,而不是各自写死。

「200 + 空 body 必须当场报错」也是刻意写死的:放过去它会在下游变成「这个博主没作品」,
把一次失败伪装成一条正常结果 —— 爬虫那条路正是这么栽的,还被翻译成「账号被封」。

测试 +11:cookie 解析、请求头成套性(含 uifid 缺失/回退)、产物键名与 store 对齐、
以及 _get 的三条失败路径(空 body / 403 带网关原话 / 正常返回)。
2026-10-10 17:12:25 +08:00

6.9 KiB
Raw Blame History

AgentOS Agent Initialization Guide

Version: 1.0.0 | Updated: 2026-06-17
Purpose: Teach AI agents how to understand and operate this system correctly.
Scope: 3-machine federated system (chengzigedeAir / 5kechengdeAir / 7kecheng)


1. Identity Card

Field Value
System AgentOS — Federated AI Exoskeleton
Machines chengzigedeAir (master), 5kechengdeAir (worker), 7kecheng (worker)
Connection Tailscale mesh network (100.x.x.x)
Dashboard Port 9988 on each machine, launchd auto-start
Git Repo ~/workbuddy-agent-os/agent-sync/ (synced to all 3 machines)

2. Python Environment (MOST IMPORTANT)

Never use system Python. The correct path is:

$HOME/.workbuddy/binaries/python/envs/agent-os/bin/python3

Or use the dynamic variable (for scripts):

$MC_PYTHON

Key packages installed:

  • camoufox 0.4.11
  • playwright 1.58.0
  • uvicorn 0.46.0
  • orjson 3.11.9

Common mistakes:

  • ❌ python3 -m mc run → ⚠️ No module named mc (needs PYTHONPATH)
  • ✅ cd $AGENT_SYNC/05_tools/07_matrix/scripts && $MC_PYTHON -m mc run ...
  • ❌ Running pip install camoufox → Already installed in venv
  • ❌ Using matrix nurture run → Legacy tool, use mc run instead

3. Command Architecture

User → Dashboard (port 9988)
  → /api/matrix/nurture/start         ← 养号执行入口
    → routes/matrix.py                  ← 自动选蓝图 + 路由到机器
      → 本机: subprocess mc run         ← 模式1
      → 远程: remote_exec exec_nurture  ← 模式2 (SSH)
        → mc run                        ← 真正的执行引擎
          → mc/engine.py                ← 身份分组 + 浏览器启动
            → cdp_connector.py          ← Camoufox 启动
              → matrix_modules/nurture/* ← 原子操作

Key commands for agents:

# 正确的养号执行(新路径,推荐)
cd $AGENT_SYNC/05_tools/07_matrix/scripts
$MC_PYTHON -m mc run --accounts=douyin_01 --blueprints=douyin_daily --rounds=10

# 查看账号列表
$MC_PYTHON -m mc account list

# 智能登录
$MC_PYTHON -m mc smart-login douyin_01

# 远程执行(通过 Dashboard API)
curl -X POST http://localhost:9988/api/federation/nurture \
  -H "Content-Type: application/json" \
  -d '{"machine":"7kecheng","accounts":["douyin_137"],"blueprints":["douyin_daily"],"rounds":10}'

4. Identity Directory Resolution

Directory naming: Identity dirs are named by phone number, NOT account ID:

identities/
├── phone_18550099569/        ← 身份目录(手机号命名)
│   ├── config.yaml
│   └── user_data/            ← 浏览器持久化数据
├── phone_15358497926/
│   ├── config.yaml
│   └── user_data/

Account-to-identity mapping: accounts.yaml links account IDs to identity dirs:

- id: douyin_136               # 账号 ID
  identity_dir: phone_18550099569   # 身份目录(映射关系)
  phone: '18550099569'
  platform: douyin
  owner_machine: 7kecheng        # 所属机器

How mc run resolves it (mc/engine.py:269):

# Step 1: Get identity_dir from account info
identity_dir = acct.get("identity_dir", acct["id"])  # → "phone_18550099569"

# Step 2: Strip "identities/" prefix if present
identity_dir = identity_dir.replace("identities/", "")  # → "phone_18550099569"

# Step 3: Resolve full path
full_path = IDENTITIES_ROOT / identity_dir  # → ~/.../identities/phone_18550099569/

❗ NEVER look for identities/{account_id}/. The directories are named by phone number.


5. Machine Routing & Blueprint Auto-Selection

Machine routing:

Each account has an owner_machine field. The API auto-routes:

  • Local (owner_machine == this machine) → subprocess.Popen
  • Remote (owner_machine != this machine) → SSH via remote_exec.py

Blueprint auto-selection:

Account Platform Default Blueprint
douyin douyin_daily
xiaohongshu xhs_daily

6. Dashboard API

The Dashboard runs on each machine (port 9988, launchd auto-start on boot).

Key API endpoints:

Endpoint Method Description
/api/identity GET Returns this machine's identity
/api/machines GET Returns all machines' status
/api/matrix/accounts GET List all accounts (federated)
/api/matrix/nurture/start POST Execute nurture (see usage below)
/api/federation/nurture POST Remote nurture execution
/api/health GET Health check

/api/matrix/nurture/start request format:

{
  "accounts": ["douyin_01", "xhs_01"],
  "rounds": 10,
  "blueprint": "douyin_daily",        // optional, auto-detect if empty
  "machine": "7kecheng",              // optional, auto-route if empty
  "dry_run": true                     // preview only, don't execute
}

7. Common Pitfalls for AI Agents

Mistake Why Fix
Using python3 Wrong Python (no packages) Use $MC_PYTHON or absolute venv path
Running matrix nurture run Legacy CLI, doesn't handle identity mapping Use mc run instead
Looking for identities/{account_id}/ Dirs named by phone number Check accounts.yaml → identity_dir field
Using --dry-run flag mc run doesn't support it Use API's dry_run: true field instead
Installing packages with pip Already in agent-os venv Use $HOME/.workbuddy/binaries/python/envs/agent-os/bin/pip3 if needed
Running mc run from wrong directory Module not found Must cd to $AGENT_SYNC/05_tools/07_matrix/scripts first
Using CDP port 9222 Legacy Chrome, now uses Camoufox Default: camoufox persistence mode

8. Quick Health Check

# Check Python venv
$HOME/.workbuddy/binaries/python/envs/agent-os/bin/python3 -c "from camoufox import AsyncCamoufox; print('OK')"

# Check mc CLI
PYTHONPATH=$AGENT_SYNC/05_tools/07_matrix/scripts $HOME/.workbuddy/binaries/python/envs/agent-os/bin/python3 -m mc --help

# Check Dashboard
curl -s --connect-timeout 3 http://localhost:9988/api/identity

# List accounts
PYTHONPATH=$AGENT_SYNC/05_tools/07_matrix/scripts $HOME/.workbuddy/binaries/python/envs/agent-os/bin/python3 -m mc account list

# Test nurture preview
curl -s -X POST http://localhost:9988/api/matrix/nurture/start \
  -H "Content-Type: application/json" \
  -d '{"accounts":["douyin_test"],"rounds":1,"dry_run":true}'

9. Environment Variables

Variable Typical Value Purpose
$AGENT_SYNC ~/workbuddy-agent-os/agent-sync Synced git repo root
$AGENT_LOCAL ~/workbuddy-agent-os/agent-local Local (gitignored) data
$MC_PYTHON ~/.../agent-os/bin/python3 Dynamic Python discovery
$HOME /Users/chengzige (or 5kecheng/7kecheng) Machine-specific home