DAILY INTEL · 20260727 · EDITORIAL

今天只留能變現的訊號

Self-evolve rank 20260727

41sources ranked
capability drill
32skills in motion
Top signal

主訊號 · Rank

  • ECC | Agent 核心架構:定義技能、本能、記憶與安全的系統化模型 | 落地:將 ECC 作為 agent 的基礎架構,定義技能與本能
  • Karpathy Skills | 減少冗贅:從 Karpathy 觀察中提煉出 LLM 編碼的行為規範 | 落地:將 CLAUDE.md 注入 agent,讓其遵循特定編碼風格
  • Graphify | 知識圖譜:將整個 codebase 轉為可查詢的知識圖,不依賴向量 | 落地:使用 /graphify 技能,讓 agent 能在圖中定位每個邊緣的意義
  • Caveman | 節省 Token:用「穴居人」風格砍掉 65% 的冗贅字數 | 落地:啟用 caveman 技能,讓 agent 用極簡的語氣溝通
  • Ponytail | 懶人資深:讓 agent 思考「最好的程式就是從未寫出」的資深風格 | 落地:啟用 ponytail 技能,讓 agent 思考如資深工程師
  • Open Design | 設計引擎:將設計從文字解放,讓 agent 生成原型、網頁與影片 | 落地:將設計任務外包給 open-design 技能
Models

雙模狀態

8080 maker 8081 checker local only

  • Web Elite v2 skill 已掛載
  • Editorial news shell 啟用
  • channel strip + hard_audit
Labs

Verify / Labs

  • 本能層:不可違反的基礎行為(如:不刪 GGUF、一任務一 skill、8081 parallel=1)
  • 技能層:可組合、可獨立啟用/停用的行為模組
  • 記憶層:區分短期(context window)與長期(file-based)
  • 安全層:定義寫入權限、刪除保護與工具邊界
  • 驗證:本能層在 10 次任務中無違反、技能層可獨立啟用/停用、記憶層在 context 溢出時正確回退、安全層保護 GGUF 不被刪除
Ship

Ship notes

  • items: 41
  • raw: D:\ai-daily-data\evolution\daily\raw_20260727.json
  • rank: D:\ai-daily-data\evolution\daily\rank_20260727.md
  • draft: D:\ai-daily-data\evolution\skill-drafts\self_evolve_20260727_050002.md
  • verify: D:\ai-daily-data\evolution\daily\verify_20260727.md

Papers & sources

今日排序 · 可掃讀

2026-07-22
OpenForgeRL: Train Harness-native Agents in Any Environment
hf_daily
src
2026-07-17
Dataset Distillation by Influence Matching
hf_daily
src
2026-07-15
Multi-Turn On-Policy Distillation with Prefix Replay
hf_daily
src
2026-07-22
SANA-Video 2.0: Hybrid Linear Attention with Attention Residuals for Efficient Video Generation
hf_daily
src
2026-07-20
FinanceComplexQA: Benchmarking Agentic Reasoning on Industrial-grade Financial Documents
hf_daily
src
2026-07-22
Self-Supervised Learning of Structured Dynamics from Videos
hf_daily
src
2026-07-22
K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs
hf_daily
src
2026-07-22
Sample-Efficient Learning from Agent Experience
hf_daily
src
2026-07-13
Color Pass-Through via Camera-Display Coupling
hf_daily
src
2026-07-21
LLMs Get Lost in Evolving User Intent
hf_daily
src
2026-07-22
Show, Don't Tell: Evaluating Spatial Cognition in Generative Pixels Rather Than LLM Text
hf_daily
src
2026-07-22
Recurrent Sinusoidal INRs for Efficient High-Fidelity Representation
hf_daily
src
2026-07-22
Visual Contrastive Self-Distillation
hf_daily
src
2026-07-21
Robostral Navigate
hf_daily
src

Skills distilled

可落地能力

UI

web-elite-2026 v2

GSAP · Lenis · Three · OKLCH · 反 AI-slop · zero-build/Vite 雙軌

News

commercial-news-ui-2026

編輯氣質 bento 情報站殼 · 8081 ≥9 才 deploy

draft

k2s_20260727_051159_context-loop-wisdom

--- name: context-loop-wisdom description: 8080/8081 雙模型自我提示循環作業系統,驅動 Maker 自主推進任務 version: 20260725 --- # context-loop-wisdom ## Purpose 當 CLI 啟動 8080 Maker 時載入。用於需要自主推進、不依賴人類微步驟的作業情境。8080 負責建(Mak

draft

ops_ai9-fast-deploy__20260727_0510

``markdown version 1.1 changelog: 注入失敗回退、驗證、一鍵指令與歸因路徑。 # AI9 快速部站 ## 伺服器指紋 - IP: 119.14.175.24 / host ai9 / Win Server 2025 / i5-9500 / 16GB - SSH: rovk@119.14.175.24 - 網域: brodyaitw.com` -

draft

_report_20260727_0510

# Skill Evolve 20260727_0510 - ops/youtube-quick-summary -> D:\ai-daily-data\evolution\skill-drafts\ops_youtube-quick-summary__20260727_0510.md - ops/ai9-fast-deploy -> D:\ai-daily-data\evolution\ski

draft

ops_youtube-quick-summary__20260727_0510

## SKILL.md version: 1.0 changelog: 8081 注入:失敗回退、驗證、一鍵指令、歸因路徑 D:/ai-daily-data/evolution/attribution/ --- name: youtube-quick-summary description: 快速擷取 YouTube 影片字幕並產摘要。所有產物寫入 Win11 D:\ai-daily-dat

Living rules

組織記憶

Living rules (auto-evolved)

20260725

Decouple harness (prompts/tools/loop) from base model training.

Capture harness-native signals (tool calls, reasoning, rewards).

Train harness layer independently of base model.

Fall back to joint training if decoupling fails.

Avoid retraining the base model when only harness changes.

Agentic Context Management