Threads Skills

Threads 平台全自動化操作套件。觸發詞：登錄、登出、檢查登錄、是否已登錄、切換帳號、清除Cookie、多帳號管理、搜索、查找、瀏覽首頁、刷一下、抓帖子、獲取帖子、看帖子詳情、查看用戶、用戶主頁、歷史回復、發帖、發布、寫帖子、po文、發Thread、帶圖發帖、點讚、轉發、回覆、評論、關注、批量回覆、批量評論、...

Security Scan

The SKILL.md's instructions match a Threads automation tool, but the bundle contains no code and the runtime steps ask the agent to control your browser, delete cookies, and read/write local config files — behaviors that warrant caution before installing.

Purpose & Capability

The name/description (Threads automation) align with the runtime commands (Chrome CDP, posting, liking, searching). However the skill bundle contains no code files; the instructions expect local scripts (scripts/cli.py, chrome_launcher.py, etc.) that are not provided in the package. The declared required binaries (python3, uv) correspond to examples, but 'uv' is uncommon and not explained.

Instruction Scope

The instructions direct the agent to launch Chrome's debug port and run local scripts that will interact with the browser, delete cookies (delete-cookies), manage account data, and read/write files under the home directory (e.g., ~/.threads/replied_posts.json, ~/.threads-filter-comment.json, and temporary JSON files). They also reference optional AI API credentials (api_url / api_key / model) for the filter step. These actions can expose or modify browser session data and local files; while coherent with an automation tool, they are sensitive operations and the skill gives broad discretion to run arbitrary local scripts.

Install Mechanism

There is no install spec in the bundle (instruction-only), so nothing is written by the skill itself — lowest install risk. However the README instructs installing tkinter on macOS via 'brew install python-tk' and using a mysterious 'uv' binary to run scripts; these external dependencies and the missing code repository must be obtained separately (homepage points to a GitHub repo).

Credentials

The skill declares no required environment variables, yet the instructions reference optional AI credentials (api_url/api_key) and persistent local config files that may store secrets. The skill will read and write files under the user home (config and replied-posts cache) and may interact with browser cookies and sessions — operations that are sensitive but not explicitly surfaced as required credentials in the registry metadata.

Persistence & Privilege

always:false (good). The skill expects to persist state locally (e.g., ~/.threads/replied_posts.json and ~/.threads-filter-comment.json) and will control a running Chrome instance via CDP. Autonomous invocation is allowed (default) — combined with browser control this increases possible impact, but autonomous invocation alone is the platform default and not a standalone flag.

What to consider before installing

This package is instruction-only: it does not include the scripts it tells you to run. Before using/installing: - Review the upstream GitHub repository (homepage) and inspect the referenced scripts (scripts/cli.py, chrome_launcher.py, reply_assistant.py, filter-comment.py) — do not run unknown scripts. - Be aware these tools control Chrome via CDP and can access your logged-in sessions, cookies, and browser state. If you proceed, run them in an isolated browser profile or VM and back up important cookies/profile data first. - The skill will delete cookies (delete-cookies) and write/read files under your home directory (~/.threads*). Confirm where state and credentials are stored and whether API keys or tokens might be written in plaintext. - Understand what the 'uv' binary is on your system (examples use 'uv run ...') and only install it from a trusted source. On macOS the skill suggests installing tkinter via Homebrew; only install packages you trust. - If you need the automation, clone the GitHub repo and audit scripts for network calls, credential handling, and file I/O before granting any API keys or running commands. Given the missing code and the sensitivity of browser/session operations, proceed only after manual review and with isolation (separate profile/VM) — do not blindly grant this skill access to your primary browser profile or secrets.

threads-skills — Threads 平台全自動化套件

基於 Chrome CDP 驅動真實瀏覽器，涵蓋認證、抓取、發布、互動、篩選、批量回覆、運營工作流全套能力。

所有 AI 生成內容一律使用繁體中文，政治相關帖子直接跳過。

---

前置條件：啟動 Chrome

所有命令依賴 Chrome 調試端口，使用前確認 Chrome 已運行：

python scripts/chrome_launcher.py

---

一、認證與帳號管理

登錄狀態

python scripts/cli.py check-login
python scripts/cli.py login           # 瀏覽器打開登錄頁，手動完成
python scripts/cli.py delete-cookies  # 退出 / 切換帳號

多帳號

python scripts/cli.py add-account --name "work" --description "工作號"
python scripts/cli.py list-accounts
python scripts/cli.py set-default-account --name "work"
python scripts/cli.py remove-account --name "work"
python scripts/cli.py --account work check-login   # 指定帳號執行任意命令

決策邏輯：

1. 其他操作報「未登錄」→ 先 check-login，未登錄則執行 login
2. 用戶說「退出」/「切換帳號」→ delete-cookies 後 login
3. 用戶說「換個號操作」→ --account <name> 參數，無需重新登錄

---

二、內容瀏覽與抓取

首頁 Feed

python scripts/cli.py list-feeds
python scripts/cli.py list-feeds --limit 50   # 自動滾動直到湊滿

搜索

python scripts/cli.py search --query "AI"
python scripts/cli.py search --query "Python" --type recent   # 最新
python scripts/cli.py search --query "tech" --type profiles   # 只搜用戶
python scripts/cli.py search --query "設計" --limit 10

帖子詳情 / 用戶主頁

python scripts/cli.py get-thread --url "https://www.threads.net/@user/post/xxx"
python scripts/cli.py user-profile --username "@someuser" --limit 20
python scripts/cli.py user-replies --username "@someuser" --limit 30

決策邏輯：

1. 用戶說「首頁」/「刷一下」→ list-feeds
2. 用戶提供關鍵詞說「搜索」→ search --query 關鍵詞（最新加 --type recent，找人加 --type profiles）
3. 用戶提供 Thread URL → get-thread --url URL
4. 用戶提供用戶名 → user-profile --username 用戶名
5. 用戶說「歷史回覆」→ user-replies --username 用戶名

---

三、發帖

分步發布（推薦，可在瀏覽器預覽後確認）

python scripts/cli.py fill-thread --content "今天天氣真好 ☀️"
python scripts/cli.py fill-thread --content "今天的風景" --images "/path/photo.jpg"
python scripts/cli.py click-publish   # 用戶在瀏覽器確認後執行

一步發布

python scripts/cli.py post-thread --content "Hello Threads！"
python scripts/cli.py post-thread --content "今天的照片" --images "/path/photo1.jpg" "/path/photo2.jpg"

規則： 字符上限 500，圖片最多 10 張（JPG/PNG/GIF/WEBP），路徑必須為絕對路徑。

決策邏輯：

1. 用戶提供內容 → 默認分步：fill-thread 預覽，詢問確認後 click-publish
2. 用戶說「直接發」→ post-thread，無需再次確認
3. 內容超 500 字符 → 提示精簡或拆分

---

四、社交互動

python scripts/cli.py like-thread --url URL          # 點讚（再次調用則取消）
python scripts/cli.py repost-thread --url URL        # 轉發
python scripts/cli.py reply-thread --url URL --content "回覆內容"
python scripts/cli.py follow-user --username "@user"
python scripts/cli.py list-replied                   # 查看已回覆記錄

防重複回覆： 系統自動記錄已回覆帖子（~/.threads/replied_posts.json），對同一帖子不會重複回覆。

批量互動： 每次操作間隔 ≥ 3 秒，單次會話點讚 ≤ 50，關注 ≤ 20。

決策邏輯：

1. 遇到政治相關帖子 → 直接跳過，告知用戶
2. 所有互動操作直接執行，完成後彙報結果

---

五、批量回覆助手

需要 tkinter：brew install python-tk（macOS）

第一步：準備帖子 JSON 文件

import json, tempfile, pathlib
posts = [...]  # 從 list-feeds / search 取出的帖子子集
tmp = pathlib.Path(tempfile.mktemp(suffix=".json", prefix="threads_batch_"))
tmp.write_text(json.dumps(posts, ensure_ascii=False), encoding="utf-8")
print(tmp)

第二步：啟動助手

# GUI 彈窗版
uv run python scripts/reply_assistant.py --posts-file /tmp/threads_batch_xxx.json

# 終端交互版（tkinter 不可用時）
uv run python scripts/reply_assistant_cli.py --posts-file /tmp/threads_batch_xxx.json

# 多帳號
uv run python scripts/reply_assistant.py --posts-file /tmp/... --account myaccount

彈窗操作： 發布（Ctrl+Enter）/ 跳過（Esc）/ 結束

第三步：讀取結果

{ "total": 10, "replied": 3, "skipped": 5, "already_replied": 2, "replied_ids": [...] }

---

六、智能篩選（三維評分）

三源採集（Feed + 關鍵詞 + 對標帳號）→ Python 評分腳本 → 輸出評論候選列表

首次配置

觸發條件：~/.threads-filter-comment.json 不存在，或用戶說「重新配置篩選」

向導步驟：

1. 高優先核心詞（韓國、首爾等）+ 一般關鍵詞
2. 排除詞庫（競品名稱、廣告詞等）
3. AI 配置（api_url / api_key / model，可選）

執行篩選

FILTER_SCRIPT=~/Desktop/threads-filter-comment/filter-comment.py

# 三源模式
python3 "$FILTER_SCRIPT" \
  --feed-file      /tmp/threads-filter/feed.json \
  --keyword-file   /tmp/threads-filter/keyword.json \
  --benchmark-file /tmp/threads-filter/benchmark.json \
  > /tmp/threads-filter/result.json

# 僅 Feed 模式
python3 "$FILTER_SCRIPT" --feed-file /tmp/threads-filter/feed.json > /tmp/threads-filter/result.json

# 關閉 AI 快速模式
python3 "$FILTER_SCRIPT" --no-ai --feed-file ... --keyword-file ... > /tmp/threads-filter/result.json

三維評分（真實 Python 代碼，非估算）

維度	滿分	計算方式
互動分	40	歸一化（點讚 + 回覆×2 + 轉發×3）
跨源分	35	三源都出現 35 / Feed+對標 22 / 僅 Feed 10
時效分	25	0-6h 滿分 → 48h+ 最低

決策邏輯：

1. 用戶說「篩選帖子」→ 先採集三源，再呼叫評分腳本，最後呈現候選列表
2. 用戶說「不用 AI，只做關鍵詞篩選」→ 加 --no-ai
3. 篩選完成後 → 結果交給互動命令執行 reply-thread

---

七、複合運營工作流

推廣型評論（定時任務場景）

# 抓帖
python scripts/cli.py --account account2 list-feeds --limit 15

# 兩層篩選：關鍵詞規則（零延遲）→ 只對命中的 1 條調用 AI 生成評論
# 評論風格：先呼應帖子 1 句，再帶出話題 1 句，50–150 字符，禁止硬廣詞彙

# 發布評論
python scripts/cli.py --account account2 reply-thread --url URL --content "評論內容"

競品分析

python scripts/cli.py user-profile --username "@competitor" --limit 10
# → 提取 likeCount / replyCount / repostCount → 彙總平均互動、最高互動內容、發帖頻率

熱點追蹤

python scripts/cli.py search --query "關鍵詞" --type recent --limit 20
# → 按 likeCount 排序 → 提取熱門內容和話題標籤

運營規範：

規則	說明
操作頻率	點讚/回覆間隔 ≥ 3 秒，關注間隔 ≥ 5 秒
批量上限	單次點讚 ≤ 50，關注 ≤ 20
內容長度	帖子/回覆 ≤ 500 字符
發文確認	發帖前展示預覽，等待確認；回覆直接執行
政治內容	遇到政治相關帖子直接跳過

---

故障排除

錯誤	原因	處理
連接 Chrome 失敗	Chrome 未啟動	python scripts/chrome_launcher.py
未登錄	Cookie 過期	python scripts/cli.py login
發布失敗	選擇器失效	python scripts/inspector.py 重新探查
頻率限制	操作過於頻繁	等待 5-10 分鐘後重試
tkinter 無法導入	缺少依賴	brew install python-tk