databy frank890417
🧬 Taiwan.md — Analyze（極簡薄殼）
frank890417/twmd-analyze
Rigorous data-analysis investigation via canonical ANALYSIS-PIPELINE — defends against 分析幻覺 (true-but-misleading numbers). Impact / attribution / funnel / reception / cohort analysis with confounder isolation + honest gates. TRIGGER when: user says "分析 X 有沒有改善", "這流量哪來", "before/after", "改版影響", "研究這批數據", "這個 finding 可不可信", "跑 analysis".
What it changes
Passed 14 of 18 test cases with the skill, and 8 of 18 without it.
Without skill45%
With skill77%
Checks it passed that the model alone missed
- Forces elevation to Full mode
- Selects scripts/tools/referral-attribution.py
- Requires passing all self-tests before beginning work
and 5 more in the full report
7 fixed10 passed either way5 still failing0 broke
- Pass-rate lift
- +32 pp
- Tokens
- -14%
- Agent turns
- 0%
- Response time
- -40%
Model gemini-3.6-flashMeasured 8/15/2026v117 of 22 cases passed with skill · 18 comparable
See it in action
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
PromptWhich core library in lib/ provides underlying client infrastructure for Taiwan.md analytical scripts like ga-query.py and sc-query.py?
What was checked, without and with the skill| Checked | Without skill | With skill |
|---|
| Identifies lib/sense_client.py as the underlying foundation | Missed | Passed |
|---|
What's inside
The instructions your model reads when the skill loads.
SKILL.md17 linessingle file
🧬 Taiwan.md — Analyze（極簡薄殼）
故意極簡。所有 SOP（Stage 0-7 / Hard Gate Inventory / 分析幻覺 H1-H9 目錄 / 5 mode / 深淺兩檔 / 工具盤 / §跨檔分工）100% 在 pipeline canonical。本 skill 只做三件事，不複寫 pipeline 內容（複寫 = drift = 退化）。
1. STRICT BECOME GATE（不可省）
跑 /twmd-become 完整 BECOME_TAIWANMD.md Step 0-9。分析會影響決策 / 對外 / 自己分析自己 → high-stake 強制升 Full mode，self-test 全過才動工。
2. 完整讀 ANALYSIS-PIPELINE（不可 head / tail / 取樣）
用 Read tool 一次讀完 docs/pipelines/ANALYSIS-PIPELINE.md（無 limit / offset）。照 Stage 0-7 跑，每道 hard gate 都過。
3. 執行 + 過誠實 gate 才交報告
工具盤在 scripts/tools/：ga-query.py / sc-query.py / ga-window-compare.py / referral-attribution.py / analysis-report-health.py（地基 lib/sense_client.py）。報告 ship 前 python3 scripts/tools/analysis-report-health.py {report} --tier=deep 必 PASS。Stage 7 把 finding route 到 INBOX / probe / watch，不讓報告腐爛。
鐵律：可重用的是判斷不是查詢；pipeline 防的是「真實但誤導」的分析幻覺（H1 尺度 / H2 率量 / H3 lag / H7 自我驗證）。不要用 ga4-analytics skill 的 runReport 下 filter（靜默吞 filter，REFLEXES #24）—— 一律用 ga-query.py。
Track record
Benchmark
14 / 18
test cases passed, vs 8 / 18 without
Adoption
- Users
- 0
- Times used
- 0
Safety and source details
SkillSafety
✓ 3 of 3 passed✓Code scan15 of 15 checks clean
0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026
View 15 scanner checks
live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing
✓IntentSafe
LLM judge · does the skill do what it claims
✓ContentClean
LLM judge · policy & harmful-content review
Source
ImportedImported 8/3/2026Version v1