Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →巴逆逆反指標分析。觸發時機：使用者要求追蹤巴逆逆、分析反指標、抓取社群貼文並推送 Telegram 時。 能力範圍：透過 CLI 抓取 Facebook 貼文、反指標邏輯分析、連鎖效應推導、Telegram 推送。 目標：由 Claude 作為分析引擎，產出直白中文的反指標分析報告。
.claude/skills/cablate-banini-tracker/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -8% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 12% | 0% |
追蹤「股海冥燈」巴逆逆（8zz）的 Facebook 貼文，由你（Claude）進行反指標分析，推送結果到 Telegram。
bash# 首次使用：初始化設定 npx @cablate/banini-tracker init --apify-token <TOKEN> --tg-bot-token <TOKEN> --tg-channel-id <ID> # 確認設定 npx @cablate/banini-tracker config
bashnpx @cablate/banini-tracker fetch -s fb -n 3 --mark-seen
輸出是 JSON 陣列，每篇貼文包含：
id / sourcetext（貼文內容）ocrText（圖片 OCR 文字，可能包含下單截圖）timestamp / url / likeCountmediaType / mediaUrl--mark-seen 會自動記錄已讀，下次不重複抓。
讀取 Step 1 的 JSON 後，進行反指標分析。分析要點：
核心邏輯（方向完全相反，不要搞混）： | 她的狀態 | 反指標解讀 | |---------|-----------| | 買入/加碼 | 該標的可能下跌 | | 被套（還沒賣） | 可能繼續跌（她還沒認輸） | | 停損/賣出 | 可能反彈上漲（她認輸 = 底部訊號） | | 看多/喊買 | 該標的可能下跌 | | 看空/喊賣 | 該標的可能上漲 |
分析原則：
連鎖效應推導：
將分析結果寫入暫存檔再推送（多行訊息用 -m 會被 shell 截斷，務必用 -f）：
bash# 寫入暫存檔後推送（推薦） npx @cablate/banini-tracker push -f /tmp/report.txt # 短訊息可用 -m npx @cablate/banini-tracker push -m "短訊息" # 純文字（不解析 HTML） npx @cablate/banini-tracker push -f /tmp/report.txt --parse-mode none
bash# 去重管理 npx @cablate/banini-tracker seen list # 列出所有已讀 ID npx @cablate/banini-tracker seen mark <id...> # 手動標記已讀 npx @cablate/banini-tracker seen clear # 清空已讀紀錄 # 查看/修改設定 npx @cablate/banini-tracker config # 顯示設定（token 遮蔽） # 手動編輯: ~/.banini-tracker.json
Facebook 每次抓取約 $0.02（Apify CU 計費）。
推送到 Telegram 時建議用以下 HTML 格式。注意：
url 欄位取得）< > & 必須轉義（< > &），避免 HTML 解析錯誤-f 推送<b>巴逆逆反指標速報</b>
日期：2026-04-10
<b>她的動態</b>
FB 16:56｜我想好明天要買啥了！大家期待嗎...
<a href="https://www.facebook.com/...">原文</a>
FB 14:30｜今天又被套了 油正二...
<a href="https://www.facebook.com/...">原文</a>
<b>反指標分析</b>
↓ 原油正二（ETF）
她：被套持有中 → 反指標：油價可能繼續跌
原因：她還沒認輸停損，底部還沒到
<b>連鎖推導</b>
油價續跌 → 原物料成本降 → 製造業毛利提升 → 電子代工股受惠
<b>建議方向</b>
關注電子代工族群，等她停損油正二那天再考慮反手做多油價
冥燈指數：8/10
<i>僅供娛樂參考，不構成投資建議</i>| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 16,127 | 5,764 | -64% | 1 | 1 | 0% | 2,395 | 1,545 | -35% | 0 | 0 | — |
case-02 | fail→fail | 12,716 | 2,586 | -80% | 1 | 1 | 0% | 1,904 | 1,516 | -20% | 0 | 0 | — |
case-03 | fail→fail | 14,291 | 5,807 | -59% | 1 | 1 | 0% | 2,243 | 2,072 | -8% | 0 | 0 | — |
case-04 | pass→pass | 16,459 | 15,898 | -3% | 1 | 1 | 0% | 2,925 | 4,482 | +53% | 0 | 0 | — |
case-05 | pass→pass | 15,476 | 9,855 | -36% | 1 | 1 | 0% | 3,082 | 3,182 | +3% | 0 | 0 | — |
case-06 | pass→pass | 17,126 | 15,894 | -7% | 1 | 1 | 0% | 2,954 | 4,256 | +44% | 0 | 0 | — |
case-07 | fail→pass | 10,059 | 2,280 | -77% | 1 | 1 | 0% | 1,763 | 1,625 | -8% | 0 | 0 | — |
case-08 | fail→pass | 14,718 | 11,484 | -22% | 1 | 1 | 0% | 2,086 | 2,967 | +42% | 0 | 0 | — |
case-09 | pass→pass | 15,488 | 8,325 | -46% | 1 | 1 | 0% | 2,221 | 2,461 | +11% | 0 | 0 | — |
case-10 | fail→pass | 14,373 | 4,319 | -70% | 1 | 1 | 0% | 2,587 | 2,006 | -22% | 0 | 0 | — |
case-11 | fail→pass | 6,290 | 1,762 | -72% | 1 | 1 | 0% | 1,100 | 1,470 | +34% | 0 | 0 | — |
case-12 | fail→pass | 8,833 | 2,955 | -67% | 1 | 1 | 0% | 1,524 | 1,700 | +12% | 0 | 0 | — |
case-13 | fail→pass | 11,195 | 1,723 | -85% | 1 | 1 | 0% | 1,919 | 1,434 | -25% | 0 | 0 | — |
case-14 | fail→pass | 9,038 | 2,049 | -77% | 1 | 1 | 0% | 1,504 | 1,527 | +2% | 0 | 0 | — |
case-15 | fail→pass | 8,570 | 2,242 | -74% | 1 | 1 | 0% | 1,654 | 1,606 | -3% | 0 | 0 | — |
case-16 | pass→pass | 16,023 | 6,131 | -62% | 1 | 1 | 0% | 2,406 | 2,195 | -9% | 0 | 0 | — |
case-17 | pass→pass | 22,537 | 14,348 | -36% | 1 | 1 | 0% | 3,186 | 3,332 | +5% | 0 | 0 | — |
case-18 | fail→pass | 19,227 | 12,303 | -36% | 1 | 1 | 0% | 2,726 | 3,106 | +14% | 0 | 0 | — |
case-19 | pass→pass | 8,767 | 4,314 | -51% | 1 | 1 | 0% | 1,579 | 2,002 | +27% | 0 | 0 | — |
case-20 | fail→fail | 13,576 | 2,181 | -84% | 1 | 1 | 0% | 2,196 | 1,566 | -29% | 0 | 0 | — |
case-21 | fail→pass | 9,283 | 2,787 | -70% | 1 | 1 | 0% | 1,721 | 1,691 | -2% | 0 | 0 | — |
case-22 | fail→pass | 18,643 | 2,807 | -85% | 1 | 1 | 0% | 3,126 | 1,695 | -46% | 0 | 0 | — |
case-23 | fail→pass | 13,291 | 2,405 | -82% | 1 | 1 | 0% | 1,984 | 1,622 | -18% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +52 percentage points is the difference between those two pass rates over the 23 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.