documentationby frank890417
🧬 Taiwan.md — Article Inbox 消化入口
frank890417/twmd-article-inbox
Add topics/material to ARTICLE-INBOX via canonical BRANCH-PIPELINE — digest observer input (theme / article / raw material) into curated candidates with priority + 連結密度 + 對比理由, then append §Pending. TRIGGER when: user says "/twmd-article-inbox X", "把 X 加進 article inbox", "幫我研究 X 放進 inbox", "X 進 inbox 消化".
What it changes
Passed 9 of 16 test cases with the skill, and 2 of 16 without it.
Without skill14%
With skill59%
Checks it passed that the model alone missed
- Searches active pending items in ARTICLE-INBOX
- Searches past completed records in ARTICLE-DONE-LOG
- Displays real-time pending entry count from ARTICLE-INBOX
and 15 more in the full report
11 fixed2 passed either way8 still failing1 broke
- Pass-rate lift
- +45 pp
- Tokens
- +2%
- Agent turns
- 0%
- Response time
- -35%
Model gemini-3.6-flashMeasured 8/15/2026v113 of 22 cases passed with skill · 16 comparable
See it in action
One of 11 test cases the skill fixed. Same model, same prompt, with and without the skill.
PromptWhat mode parameter must be passed to the become command when initiating an article inbox processing session?
What was checked, without and with the skill| Checked | Without skill | With skill |
|---|
| Specifies mode=write when running the initial become tool command | Missed | Passed |
|---|
What's inside
The instructions your model reads when the skill loads.
SKILL.md28 linessingle file
🧬 Taiwan.md — Article Inbox 消化入口
🚨 STRICT BECOME GATE — 第一動作不可省略
Before anything else：跑 /twmd-become write 完整走 BECOME_TAIWANMD.md Step 0-9（write mode 會載 ARTICLE-INBOX §P0/P1 視野，dedup 需要）。
✅ BECOME ack: mode=write / ARTICLE-INBOX pending 數=<inbox-signal 即時讀> / Q14=PASS
Pipeline
嚴格完整讀取並執行 docs/pipelines/BRANCH-PIPELINE.md v2.1 Stage 0-5。不消化＝不 append：這條 skill 的存在理由是「進 inbox 的每條 entry 都經過分支分析」，raw 主題直接塞 inbox 是本 skill 的反面。
Mode 判斷交 Stage 0（single-article / broad-theme / hybrid；觀察者丟素材而非主題時走 §素材 intake 前處理）。
Inline hard gate（最常被偷吃步的四條，違者 = 任務 incomplete）
- Agent 落檔 claim 必驗：broad-theme spawn 的每個 research agent 回報後，主 session 必跑
ls + wc -l reports/research/{YYYY-MM}/...，檔案不存在就補寫或重派。反例：Stage 1 research agent 曾 self-report「已落檔」但檔案不存在（feedback_agent_writefile_hallucination，已升 REWRITE §13）
- 中文 verbatim + 三源：agent prompt 必含中文 query、禁英文回譯、重要事實 ≥2-3 獨立來源（REFLEXES #23 毒樹果實鏈 / #16）
- Dedup 三查：append 前 grep ARTICLE-INBOX pending + ARTICLE-DONE-LOG + knowledge/ baseline 三處。反例：只查 pending 的年代，已 ship 主題被重複推薦進 inbox，變 inbox 幽靈
- Entry 格式以 ARTICLE-INBOX §Entry Schema 為準：Requested 欄寫「YYYY-MM-DD by 觀察者/branch-analysis (session X)」，含對比理由與連結密度
收官
commit + push（worktree 走 semiont-worktree.sh ship）→ memory file 必含：BECOME ACK + mode 判定 + N candidates appended（含 P0/P1/P2 分佈）+ dedup 三查結果 + Handoff 三態。
完整 SOP：docs/pipelines/BRANCH-PIPELINE.md
Track record
Benchmark
9 / 16
test cases passed, vs 2 / 16 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
Importedfrank890417 / taiwan-md
.claude/skills/twmd-article-inbox/SKILL.md
Imported 8/3/2026Version v1