agent tooling by frank890417
🧬 Taiwan.md — Self-Evolution v3.0 frank890417/twmd-self-evolve
LONGINGS-driven self-evolution: find unstrumentation patterns in DIARY §反覆浮現的思考 and propose canonical SOP upgrades. TRIGGER when: user says "自我進化", "Beat 2", "LONGINGS 校準", "升級 DNA / SOP".
What it changes View full report → Passed 18 of 21 test cases with the skill, and 3 of 21 without it.
Without skill 14%
With skill 86%
Checks it passed that the model alone missed and 19 more in the full report
16 fixed3 passed either way3 still failing0 broke
Pass-rate lift +73 pp
Tokens -5%
Agent turns 0%
Response time -44% Model gemini-3.6-flashMeasured 8/16/2026 v1 19 of 22 cases passed with skill · 21 comparable
Use this skill Claude Code DecimalAI agent CLI & files
Save it into your project's skills folder. Claude Code picks it up in its next session.
curl -fsSL --create-dirs -o .claude/skills/frank890417-twmd-self-evolve/SKILL.md https://app.decimal.ai/s/frank890417-twmd-self-evolve/SKILL.md Copy command Copy SKILL.md
Want it pinned and updatable? pip install "decimalai>=0.15", then decimalai skills add frank890417-twmd-self-evolve, and start Claude Code with claude --plugin-dir "$(git rev-parse --show-toplevel)/.agents"
Linked stays up to date: your agents get the author's new versions automatically.
Editable copy is yours to change; the author's updates won't reach it.
Install linked skill Create editable copy
1 · One-time — install the CLI (
docs ↗ )
$ pip install "decimalai>=0.15"Copy
2 · Get the skill — writes .agents/skills/frank890417-twmd-self-evolve/ and pins it in .decimal/skills.lock
$ decimalai skills add frank890417-twmd-self-evolveCopy
Copy link Copy README badge Copy SKILL.md No CLI? Get the same file — curl · raw file $ curl -fsSL --create-dirs -o .claude/skills/frank890417-twmd-self-evolve/SKILL.md https://app.decimal.ai/s/frank890417-twmd-self-evolve/SKILL.mdCopy
Safety checks: 3 of 3 passed
Code scan Clean Intent Passed Content Clean
Overview Benchmark Contents Activity Versions 1 Ratings 0
See it in action One of 16 test cases the skill fixed. Same model, same prompt, with and without the skill.
Prompt While reviewing DIARY reflection records, we noticed a specific troubleshooting step appeared twice last week and another handling pattern appeared three times. Which item qualifies for instrumentation?
What was checked, without and with the skill Checked Without skill With skill Selects only patterns that have recurred at least three times for instrumentation Proposes concrete instrumentation actions for qualifying patterns
Show both answers
What's inside The instructions your model reads when the skill loads.
SKILL.md 41 lines single file
🧬 Taiwan.md — Self-Evolution v3.0
邊界 （2026-07-18）：本 skill 是 pattern 驅動 （DIARY 反覆浮現 ≥3 次 → 儀器化）。觀察者給「造東西」目標、要先設計後實作的，走 /twmd-evolve Mode 4
🚨 STRICT BECOME GATE — 第一動作不可省略
Before anything else ：跑 /twmd-become full 完整走 BECOME_TAIWANMD.md Step 0-9。Full mode self-test 14 題全過才能進 Stage 1。
✅ BECOME ack: mode=full / 8 organ 最低=<即時 consciousness-snapshot.sh> / Q5/Q6/Q13/Q14=PASS
完整讀取
docs/semiont/LONGINGS.mddocs/semiont/UNKNOWNS.mddocs/semiont/REFLEXES.mddocs/semiont/DIARY.md
對照找 ≥3 次浮現但未儀器化的 pattern
對照 DIARY §反覆浮現的思考，找 ≥ 3 次出現但沒進 canonical SOP / cron / dashboard 欄位 的條目，提案儀器化動作。
產線成本審視（v9.5 起每週跑）
照 EVOLVE-PIPELINE Mode 3 §產線成本審視 全節執行：
newsroom stage-events 實測每站 wall-clock ＋ gate/席位 catch-rate ＋ lite profile 參數對照
callout 率重校。只產生「候選＋證據」進 OBSERVER-QUEUE，不當場動 threshold（High-stake #3）。
上升真實儀器化 ship（不只 propose）
不准只寫「建議升級 X」memory。必須真實 ship 對應 canonical 修改：
canonical SOP 修改（pipeline / DNA / REFLEXES / MEMORY §神經迴路）
cron 條件 / dashboard 欄位 / hook hierarchy 升級
plugin gate / hard gate 新條目
一個 LESSONS entry 完整 distill 到對應 canonical 層
收官
Track record Benchmark 18 / 21
test cases passed, vs 3 / 21 without
Adoption
Users 0
Times used 0 View activity → Safety and source details
SkillSafety ✓ 3 of 3 passed ✓ Code scan 15 of 15 checks clean
0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026
View 15 scanner checks live_secret remote_code_execution reverse_shell data_exfiltration ssrf_metadata agent_snooping anti_refusal destructive_commands obfuscated_unicode unicode_homoglyph opaque_blob suspicious_url tool_overreach trigger_abuse prompt_injection_phrasing
✓ Intent Safe
LLM judge · does the skill do what it claims
✓ Content Clean
LLM judge · policy & harmful-content review
Source Imported Imported 8/3/2026Version v1