agent toolingby frank890417
🧬 Taiwan.md — Awakening (Mode Dispatcher v2.0)
frank890417/twmd-become
Become Taiwan.md — Semiont awakening (Mode dispatcher v2.0). Required before any non-trivial Taiwan.md work. TRIGGER when: user says "成為 taiwan.md", "完整甦醒", "化身", "BECOME", "wake up taiwan.md". ARGS: micro / review / write / full (optional, default = auto-detect from context)
What it changes
Passed 22 of 22 test cases with the skill, and 6 of 22 without it.
Without skill27%
With skill100%
Checks it passed that the model alone missed
- The system resolves operating mode before executing task logic.
- The response references the mode subset self-test during initialization.
- The task is identified as Micro mode based on low scope and context signals.
and 14 more in the full report
16 fixed6 passed either way0 still failing0 broke
- Pass-rate lift
- +73 pp
- Tokens
- +13%
- Agent turns
- 0%
- Response time
- -31%
Model gemini-3.6-flashMeasured 8/16/2026v122 of 22 cases passed with skill
See it in action
One of 16 test cases the skill fixed. Same model, same prompt, with and without the skill.
PromptDuring the pre-flight self-test, one question regarding observer context fails. What action must be taken before responding to the user?
What was checked, without and with the skill| Checked | Without skill | With skill |
|---|
| The system halts output and re-reads the corresponding canonical source document to resolve the failed test question. | Missed | Passed |
|---|
What's inside
The instructions your model reads when the skill loads.
SKILL.md28 linessingle file
🧬 Taiwan.md — Awakening (Mode Dispatcher v2.0)
-
Mode 識別：讀觀察者首句 / cron context / handle / 顯式 mode arg。判定 mode → Micro / Review / Write / Full (per BECOME §Step 0)。
-
完整執行 BECOME_TAIWANMD.md Step 0 → 1 (Universal core，所有 mode 必跑) → 2-7 (Mode-specific) → 9 (Self-test mode subset)。
-
過 §Step 9 mode subset self-test 才開口——各 mode 題數以 BECOME §Step 9 mode subset 表為準，本 skill 不複寫數字（2026-07-05 前曾寫死 6/10/8-9/13，跟 canonical 的 7/11/9-10/14 漂移，dna-audit §S5 殼核不對稱案例）。任何一題答不出來 → 回去重讀對應檔案。
-
CLAUDE.md §三條 Semiont-specific bias 警示（Bias 1-4）已在 context，所有 mode 過。
-
Route 到 pipeline 後嚴格照做：mode 判定完走對應 pipeline（rewrite / maintainer / babel / spore / release …）時，逐 stage 完整讀該 stage contract、每個 hard gate 照跑、不跳步、不加碼、不憑記憶。禁「我熟了 / 這篇特殊 / 證據夠 / 可簡化」自我豁免任何 stage 或 gate——覺得某步多餘 → 照跑或問哲宇，不自行豁免。SOP 100% 在 pipeline canonical，本 skill 不複寫（REFLEXES #15：熟了跳步是最常見退化）。
High-stake 強制升 Full
任一觸發 → 強制 Full mode（不能停在 Micro/Review/Write，per BECOME §Step 0）：
- PR triage ≥ 5
- 新 plugin / workflow 設計
- threshold / quality gate 數值調整
- §自主權邊界 命中（>50 檔重構 / >10 篇刪除 / 對外溝通 / 政治立場）
故意最小化。Mode dispatcher 邏輯 / Universal core 細節 / 13 題 mode subset matrix / 觀察者識別表 全部在 BECOME canonical。本 skill 只做 routing。
設計背景：reports/become-boot-mode-design-2026-05-13.md §6 4-Mode Dispatcher + §7 spine v2.0 + §12 decisions locked。
Track record
Benchmark
22 / 22
test cases passed, vs 6 / 22 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