AK體 · 數據驅動的 Threads 寫文決策系統。用你的歷史貼文、演算法與社媒心理學，協助選題、起草、發文前診斷、表現預估與復盤。Data-driven Threads writing advisor — topic selection, drafting, diagnosis, prediction & review based on your own post history.
# Add to your Claude Code skills
git clone https://github.com/akseolabs-seo/AK-Threads-boosterGuides for using ai agents skills like AK-Threads-booster.
Last scanned: 5/30/2026
{
"issues": [],
"status": "PASSED",
"scannedAt": "2026-05-30T15:31:39.828Z",
"npmAuditRan": true,
"pipAuditRan": true
}No comments yet. Be the first to share your thoughts!
30 days in the Featured rail
ak-threads-booster remains the internal skill id for routing and installation.
Use this as the single entry point for the AK Threads workflow.
This skill is for creators who want to pick stronger topics, write posts with clearer upside, and improve over time using their own Threads history.
It is not a viral-post guarantee engine. It is a decision system:
Classify the user's request first, then open and follow one primary module unless the task clearly needs a short sequence.
Each sub-skill is located via Glob **/skills/<name>/SKILL.md so resolution works regardless of where the plugin is installed. Do not assume an absolute path or environment variable.
**/skills/setup/SKILL.md**/skills/refresh/SKILL.md**/skills/analyze/SKILL.md**/skills/draft/SKILL.md**/skills/predict/SKILL.md**/skills/review/SKILL.md**/skills/topics/SKILL.md**/skills/voice/SKILL.md**/skills/panel/SKILL.md**/skills/update/SKILL.mdthreads_skill_learnings.log misses into rule edits) -> Glob **/skills/optimize/SKILL.mdthreads_daily_tracker.json is missing, do not pretend the work is data-backed. Ask for fallback history or use the setup path./analyze. /analyze gives pointed diagnosis and preserves the user's format; it does not rewrite the post. Route to /draft only when the user has no existing text and wants something generated from a topic.brand_voice.md is a composition driver only in /draft. Every other module treats it as observation-only — for flagging drift, never for rewriting the user's submission toward a voice template./draft or /analyze with this offer unless the user asks about maintenance.Look in the working directory for:
threads_daily_tracker.json - canonical machine-readable trackerstyle_guide.md - produced by /setupconcept_library.md - produced by /setupbrand_voice.md - produced by /voice, referenced by /draftposts_by_date.md - human-readable archiveposts_by_topic.md - human-readable topic indexcomments.md - human-readable flat comment logthreads_freshness.log - audit log for /draft and /topics freshness gates, read by /reviewthreads_refresh.log - audit log for /refresh runs, read by /reviewLow-token runtime also looks for derived compiled memory in compiled/:
compiled/account_wiki.mdcompiled/account_state.mdcompiled/personal_signal_memory.mdcompiled/next_move_queue.mdcompiled/post_feature_index.jsonlcompiled/cluster_wiki.jsoncompiled/exemplar_bank.mdcompiled/recent_window.mdcompiled/voice_fingerprint.mdcompiled/voice_fingerprint.jsonThese files are runtime caches only. threads_daily_tracker.json remains the source of truth. If compiled memory is missing, stale, or contradicts the tracker, fall back to the tracker and recommend rebuilding compiled memory.
If legacy Chinese companion filenames already exist, treat them as equivalent companion files instead of forcing a rename.
If only the tracker exists, continue in tracker-only fallback mode when the chosen module allows it. If the tracker is missing, do not pretend the work is data-backed - ask for fallback history or use /setup.
This main SKILL.md declares only the read-only tools it actually uses (Read, Glob, Grep). Each sub-skill declares its own allowed-tools in its frontmatter; some require more:
/draft adds Write, WebSearch, WebFetch/review adds Write, Edit/voice, /setup, /refresh each extend the surface as needed/update uses git through shell commands and may create an opt-in Codex automation when the user explicitly accepts weekly auto-updateWhen auditing permissions, inspect the union of all sub-skill frontmatters, not just this file.
Any sub-skill that writes to threads_daily_tracker.json, style_guide.md, concept_library.md, brand_voice.md, or threads_booster_config.json must follow templates/FAILSAFE.md (backup + atomic rename + keep last 5 backups). Append-only logs (threads_freshness.log, threads_refresh.log, threads_skill_learnings.log) follow the append-only rules in the same file.
Compiled memory files under compiled/ are rebuilt views, not hand-edited state. Rebuild them with scripts/build_compiled_memory.py after tracker-changing runs.
Red-line (R) and signal (S) definitions live in knowledge/_shared/red-lines.md — the single source of truth for both /analyze and /draft. Do not inline R-lists in sub-skill SKILL.md files.
Next-post direction lives in knowledge/_shared/next-move-engine.md. Any "next move" recommendation must pass red-line filtering first, name the S signal it is trying to strengthen, and avoid becoming a formula bank or case-study imitation.
Runtime depth, compiled-memory behavior, and output-mode defaults live in knowledge/_shared/runtime-budget.md. In lite and standard, use knowledge/cards/* before full knowledge/*.md files.
If runtime.token_mode is absent or "ask", ask the user to choose 低 token 版 or 高 token 版 before heavy reading. The question must clearly state the tradeoff: low token is faster and cheaper but less detailed; high token is deeper but slower and more expensive.
Compound loop schema — threads_skill_learnings.log — is defined in knowledge/_shared/compound-log-format.md. /review writes misses to it; /optimize reads it, proposes rule changes, and appends supersedes entries when the user approves edits. No other sub-skill touches the log.
When guidance conflicts, use this order:
SKILL.md — routing and global disciplineskills/<name>/SKILL.md — module-specific workflowknowledge/_shared/*.md — definitions referenced by multiple sub-skills (red-lines, config, discovery, principles, runtime-budget, compound-log-format)knowledge/*.md — deeper knowledge bases (psychology, algorithm, ai-detection, data-confidence, chrome-selectors)templates/* — shape templates (tracker, style-guide, concept-library, FAILSAFE)Rules earlier in this list win. If a sub-skill rule contradicts the main SKILL.md (e.g. routes /analyze requests to /draft), the main SKILL.md wins and the sub-skill is the drift — fix the sub-skill.
A known cross-sub-skill conflict and its resolution:
brand_voice.md usage. /draft treats it as a composition driver. Every other sub-skill treats it as observation-only. This is stated here (Routing Rules #7) and re-stated in each sub-skill's Scope section. If those ever disagree, this file wins.AK-Threads-Booster 是這個 skill 的內部代號與安裝 id。
AK-Threads-Booster 是一套給 Threads 創作者用的 AI skill 系統。
它不是要幫你亂寫一堆貼文，而是幫你把「選題、起草、分析、預測、復盤」變成一套有資料依據的工作流，讓你更容易發出值得被分享、收藏、討論的內容。
如果你平常的痛點是這些：
這套 skill 就是為這些問題設計的。
它不保證爆文。
它做的是讓你用自己的歷史資料，提高每一次發文決策的品質，讓「更有擴散機會」這件事變得比較可複製。
2.0 的重點是把 AK體從單次發文輔助，升級成更完整的 Threads content operating system。
/voice 先用本地腳本建立聲音指紋，再蒸餾核心信念、判斷張力、反 voice 禁區和 /draft 可直接使用的作戰包。/update：可以檢查新版，也可以在用戶同意後開啟每週自動檢查；只做 clean repo fast-forward，不覆蓋本地修改。AGENTS.md / SKILL.md / agents/openai.yaml。/topics 不是單純丟熱門題給你。
它會一起看：
也就是說，它不只是找「熱門」，而是找「對你這個帳號來說，現在更值得發」的題目。
/draft 會根據：
brand_voice.mdstyle_guide.md來起草一篇比較接近你語感的內容。
而且它不是拿到題目就直接寫：
這些對話功能都是開關式的——第一次會問你要不要開，也可以設成 always on / always off，存在 threads_booster_config.json。想要快就快，想要深就深。
/analyze 是這套 skill 的 decision layer。
它會看：
這樣你在按下發送前，不是只靠感覺。
/predict 會用相似歷史貼文幫你估 24 小時的可能區間，讓你不要因為單篇波動就誤判。
/review 會把實際表現、預測偏差、風格訊號再寫回 tracker。
這點很重要，因為很多工具只會給你建議，不會讓系統越用越準。這套 skill 的重點就是把學到的東西留下來。
/refresh 可以更新 threads_daily_tracker.json：
你不用每次都自己慢慢補資料。
/voice 會把歷史貼文變成更可執行的 Brand Voice：
/draft 可以直接用的 Quick-Reference Pack 和 Forbidden Zone這讓 /draft 不只模仿句型，而是更接近你的內容基因。
/panel 會開啟本地 visual panel，讓你不用翻 JSON 也能看：
面板本身不會呼叫 AI。你可以先看資料，再決定要不要把某篇交給 agent 分析。
/update 可以檢查 GitHub 上是否有新版。
它只會在本地 repo 乾淨、可以 fast-forward 時更新；如果你有本地修改、local-only commits 或衝突，它會停下回報，不會自動覆蓋。
這套 skill 特別適合：
如果你現在還完全沒有歷史資料，它也可以用，但前期的判斷會比較弱。這套系統的價值，會隨著你的資料累積而變強。
跑完 /setup 之後，工作目錄通常會有：
threads_daily_tracker.jsonstyle_guide.mdconcept_library.mdbrand_voice.md（如果有跑 /voice）compiled/account_wiki.mdcompiled/account_state.mdcompiled/next_move_queue.mdcompiled/post_feature_index.jsonlcompiled/voice_fingerprint.md / .json（如果有重建 compiled memory 或跑新版 /voice）posts_by_date.mdposts_by_topic.mdcomments.md其中最重要的是 threads_daily_tracker.json。
其他檔案都是圍繞這份 tracker 產生的 companion 或 runtime cache。tracker 永遠是 source of truth，compiled memory 可以重建，不需要手動改。
/setup
/voice
先把歷史資料整理好，再把 Brand Voice 建起來。新版 /voice 會先用本地腳本做 voice fingerprint，再讓 AI 蒸餾認知層、反 voice 禁區和 /draft 作戰包。
/voice 產出的 brand_voice.md 是參考初稿，不是定稿。LLM 從外部看你的貼文一定會漏東西。建議：
/voice 會保留你改過的內容，不會覆蓋掉/draft 會把 Manual Refinements 當硬約束讀，優先級高於其他章節；接著讀 Cognitive Core、Quick-Reference Pack、Anti-Voice 和 Voice Fingerprint。
/topics
/draft
/analyze
這是最實用的一組流程：
/topics 找題/draft 起草/analyze 發文前檢查/predict
/review
這樣系統會慢慢知道：
/panel
或在本機執行：
python scripts/panel_server.py --open
面板適合在寫文前先掃 30 秒，決定這次要不要進入 /topics 或 /analyze。
你可以用這些方式建立資料：
API 不是必須，但如果你有 API，更新會輕鬆很多。
AK-Threads-Booster 是一套以你的 Threads 歷史資料為核心的內容決策系統。
它的重點不是自動亂生文，而是幫你：
把這個 GitHub repo 給你的 agent：
https://github.com/akseolabs-seo/AK-Threads-booster
支援 skill / repo instructions 的 agent 可以直接讀 AGENTS.md 或 SKILL.md，再依你的指令進入 /setup、/voice、/topics、/draft、/analyze 等模組。
支援 OpenAI/Codex-style discovery 的環境也可以讀 agents/openai.yaml 作為 UI metadata。
也可以手動 clone：
git clone https://github.com/akseolabs-seo/AK-Threads-booster.git
再依你使用的工具，把這個 repo 放到對應的 skill / agent instructions 目錄即可。
安裝後可以用 /update 檢查 AK-Threads-Booster 有沒有新版。
/update 會主動問你要不要開啟每週自動檢查更新。開啟後只有在本地 repo 乾淨、可以 fast-forward 時才會更新；如果你有本地修改、local commits 或衝突，它會停下回報，不會覆蓋你的東西。
AK-Threads-booster/
|- SKILL.md
|- AGENTS.md
|- agents/
| |- openai.yaml
|- skills/
| |- setup/SKILL.md
| |- refresh/SKILL.md
| |- analyze/SKILL.md
| |- draft/SKILL.md
| |- predict/SKILL.md
| |- review/SKILL.md
| |- topics/SKILL.md
| |- voice/SKILL.md
| |- panel/SKILL.md
| |- update/SKILL.md
|- knowledge/
| |- _shared/
| |- psychology.md
| |- algorithm.md
| |- ai-detection.md
| |- data-confidence.md
| |- chrome-selectors.md
|- scripts/
| |- fetch_threads.py
| |- parse_export.py
| |- build_compiled_memory.py
| |- build_voice_distillation.py
| |- check_skill_update.py
| |- panel_server.py
| |- update_snapshots.py
| |- update_topic_freshness.py
| |- render_companions.py
|- panel/
|- templates/
|- examples/
MIT License. See LICENSE.