Agent Workflows Topic Map

這是 AI agent 工作流、harness、loop、subagent 與動態編排的主題導覽。

Core concepts

• ai-agent — agent 的基本定義
• personal-ai-agent-seven-layer-architecture — 個人 AI 助理的七層 harness 骨架
• personal-general-ai-assistant — agentic coding 之後，推理、工具使用與個人化匯流的助理產品形態
• agent-economy — 個人與企業 agent 代替使用者協調供需、溝通與交易的市場結構
• agentic-commerce-protocol-landscape — agentic commerce 的協定、介面與場域整合地圖
• harness-engineering-for-ai-coding — 讓 agent 的輸出可驗證、可否決、可修正
• agentic-engineering-patterns — 以測試、理解、Git、subagent 與 PR evidence 把低產碼成本轉成可維護品質
• agent-instruction-files — 專案層級 agent 指令檔：只放不寫就會錯的內容，紅線交給 hooks
• agent-skills — 可重複工作流程的封裝、載入與任務級評估
• agent-experience — 讓 CLI、API、文件與產品介面能被 agent 低成本探索、操作與修復
• claude-learning-map — 以工作場景選擇 Claude Chat、Cowork、Code、Skills 與 Claude Code 的學習路徑
• loop-engineering — 把任務變成 discover / execute / verify / iterate 的閉環
• agent-framework-selection — 在 Deep Agent 與基礎框架之間選型
• unix-style-agent-architecture — 以 shell、檔案、Markdown 與 cron 組成最小 agent runtime
• model-interface-agent-orchestration — 將 agent orchestration 下沉到 provider 的 model endpoint
• agent-sandbox-architecture — agent sandbox 的隔離模式、權限邊界與 execution safety
• frontier-model-release-governance — 前沿模型能力、safeguards 與 trusted access 的發布治理
• identity-aware-connectivity-platform — 以 identity、resource、policy／time、action 與 trace 管理 agent connectivity
• self-improving-harness — 讓 harness 改動進入可驗證、可回滾的改進迴圈
• dynamic-agent-workflows — 根據工具回饋動態調整策略
• graph-engineering — 用節點、交棒、狀態與預算組織多個 agent／loop
• cognitive-load-agent-orchestration — 用 CLT 判斷多 agent 任務拆分、spec 精度與 memory 外部化
• agent-ready-data-governance — 企業資料與現場知識整理成 agent 可用的流程底座，從單機檔案、雲端系統到可授權的跨系統資料通道
• forward-deployed-engineering — FDE：從現場流程、系統整合到客戶接手的部署模式
• ai-fitness-and-enterprise-ai-maturity — 用基礎能力、應用能力與財務績效理解企業 AI 成熟度
• wearable-health-foundation-models — domain foundation model 供 agentic classroom 適配並 grounding Personal Health Agent
• chain-of-evidence-autonomous-research — 用 evidence chain、claim verifier 與 integrity audit 約束 autonomous research agent
• discovery-loop — 把「實驗、觀察、修正」的 research loop 推向科學與工程自動化
• planetary-prediction-engine — LLM 編排地理空間資料選擇、整理、AutoML 與預測的研究 workflow
• time-series-foundation-models — 預訓練時間序列模型的 multivariate、covariate、quantile 與評測契約
• data-science-agents — 異質資料、程式執行與 verifier loop 驅動的資料科學 agent
• ai-native-enterprise-governance — AI teammate 的授權邊界、owner、sandbox、observability 與 workflow KPI
• agentic-ai-cost-management — 以 task-level cost、效率前沿、模型路由與漸進式預算治理管理 agent 成本
• agentic-search-tool-curation — 依 query understanding、結構化條件與 vector fallback 策展 agent 檢索工具
• document-parsing-first-rag — 先檢查文件結構與 parser failure，再選 retrieval／agentic loop
• ai-code-validation-bottleneck — coding agent 產碼暴增後的規格、審查與整合驗證瓶頸
• code-abundance-product-taste — Codex 讓人人都能建東西後，產品取捨、時機與整合變成稀缺能力
• ai-native-engineering-organization — coding 變便宜後，工程組織如何重設 norms、驗證與協作
• ai-collaboration-compounding-workflow — 讓個人與團隊 AI 協作越用越順的 context / workflow 複利模式

Comparisons

• cloud-agent-vs-localhost-agent — cloud delegation 與 localhost pairing 的環境、權限與採用邊界
• harness-engineering-vs-loop-engineering — harness 與 loop 的分工
• agent-teams-vs-dynamic-workflows — agent teams、subagents、dynamic workflows 的差異
• skills-vs-mcp — skills、CLI / script 與 MCP server 的分工邊界

Operating layer

• aiark-loop-engineering — AI Ark wiki 的常駐 watch / ingest / verify / synthesize loop
• agent-trace-observability — 用 trace 校準 agent 行為
• prompt-debugging-eval-checklist — 用 eval checklist 修提示詞

Raw sources

• raw/articles/bnext-salesforce-claudeforce-dumb-pipe-2026-09-04.md — Salesforce／Anthropic Claudeforce、agent 入口、資料／工作流護城河、MCP 可攜性與 SaaS「笨水管」風險

• raw/articles/bnext-ai-coding-costs-at-scale-2026-09-04.md — Databricks AI coding 成本管理、效率前沿、model／harness flexibility、路由、預算閘門與 token overhead

• raw/articles/bnext-gpt-6-astra-critical-cybersecurity-launch-2026-09-04.md — GPT-6 Astra 的 Critical cyber threshold、harness／benchmark 邊界、monitorability 與 Trusted Access／Daybreak release gate

• raw/articles/bnext-meta-muse-spark-1-3-2026-09-03.md — Muse Spark 1.3 的 agentic／coding 評測、effort／task cost、tool-loop 效率與 max reasoning release gate

• raw/articles/bnext-gemini-3-8-flash-cyber-2026-09-03.md — Gemini 3.8 Flash／Cyber 的 reasoning／tool loop、task-level cost、資安評測與 Fairwind trusted-access 邊界

• raw/articles/bnext-anthropic-fable-5-1-mythos-5-1-2026-09-02.md — Fable 5.1／Mythos 5.1 的 model-harness-fit、task-level cost、API migration 與 safeguards release path

• raw/articles/bnext-dhh-ai-programming-lex-fridman-2026-09-01.md — DHH 的 problem-first prompting、agentic engineering、產品品味、既有 codebase 整合與開源維護邊界

• raw/articles/bnext-openclaw-2-0-release-install-guide-2026-09-01.md — OpenClaw 2.0 的 guided setup、Control UI、Skills／memory、Cloud Session 與 trusted-team collaboration 邊界

• raw/articles/networkworld-tailscale-connectivity-platform-2026-08-31.md — Tailscale 從 WireGuard VPN 延伸至 identity-aware connectivity、Aperture、PAM、DNS filtering、Tailcat 與 programmable tailnet

• raw/papers/pwc-ai-performance-global-taiwan-2026-07-17.md — AI fitness、企業 AI 成熟度、治理與規模化研究報告

• raw/articles/raymondhouch-ai-agent-7-layer-architecture-2026-06-30.md

• raw/articles/aihao-harness-engineering-deep-agent-capabilities-2026-06-26.md

• raw/articles/ihower-harness-loop-engineering-2026-06-26.md

• raw/articles/aihao-harness-engineering-mid-run-injection-2026-06-26.md

• raw/articles/aihao-harness-engineering-outer-loop-2026-06-26.md

• raw/articles/aihao-harness-engineering-self-improving-2026-06-26.md

• raw/articles/aihao-harness-engineering-agent-frameworks-2026-06-26.md

• raw/articles/aihao-cognitive-load-agent-orchestration-2026-06-17.md

• raw/articles/bnext-wistron-ai-factory-agent-smart-manufacturing-2026-06-26.md

• raw/articles/bnext-szs-ai-agents-in-house-development-2026-08-01.md

• raw/articles/bnext-ai-native-enterprise-transformation-governance-2026-06-16.md

• raw/articles/aihao-ai-enabled-ai-first-ai-native-2026-08-22.md

• raw/articles/bnext-claude-code-management-fiona-fung-2026-06-26.md

• raw/articles/bnext-egenthub-fde-2026-08-27.md — EgentHub 的 FDE、超級個體、跨模型選擇與客戶能力擴散

• raw/articles/bnext-kdan-hr-ai-recruiting-automation-2026-08-27.md — AI 招募流程自動化、流程變體、漸進採用與人類高信任節點

• raw/articles/bnext-orra-ai-employee-training-4-stages-2026-08-28.md — ORRA AI 員工的招募、訓練、考核、上工後管理與人類判斷邊界

• raw/articles/bnext-claude-cowork-scheduled-tasks-2026-08-28.md — Claude Cowork 排程、獨立 session、遠端／本機 runtime 與結果 review 邊界

• raw/articles/bnext-claude-cowork-built-in-browser-2026-08-28.md — Claude Cowork 內建瀏覽器、browser identity、Claude in Chrome 分工與 prompt injection 邊界

• raw/articles/bnext-openai-codex-growth-legal-108x-engineer-5x-2026-08-28.md — OpenAI Enterprise Signals、Codex 跨職能採用、企業 agent 執行化與 Token 指標界線

• raw/articles/bnext-glassdoor-employer-ai-2026-08-31.md — Glassdoor 職場 AI 評價、世代／性別差異、工作安全感與企業採用治理訊號

• raw/articles/bnext-chatgpt-work-sign-in-websites-2026-08-30.md — ChatGPT Work cloud browser 登入網站、secure sign-in、背景執行與人工確認邊界

• raw/articles/bnext-zeabur-env-var-leak-api-key-stolen-2026-08-31.md — Zeabur 平台 secrets、credential blast radius、query／export trace、憑證輪替與支出上限邊界

• raw/articles/bnext-openai-cursor-model-supply-2026-08-31.md — OpenAI／Cursor 模型供應、控制權變更條款、第三方 coding harness 連續性與 fallback 邊界

• raw/articles/bnext-claude-memory-chat-cowork-unified-2026-08-30.md — Claude Chat／Cowork 記憶互通、Topics 編輯控制、敏感主題與 cloud／local runtime 邊界

• raw/articles/bnext-openai-codex-taste-over-coding-2026-06-29.md

• raw/articles/bnext-spotify-ai-engineering-validation-loops-2026-06-30.md

• raw/articles/bnext-meta-bans-claude-code-codex-2026-06-30.md

• raw/articles/aihao-ai-product-management-2026-05-18.md

• raw/articles/bnext-google-deepmind-ai-agent-inference-2026-06-26.md

• raw/articles/bnext-meta-ai-agent-economy-2026-06-29.md

• raw/articles/bnext-agentic-commerce-department-store-2026-06-29.md

• raw/articles/google-research-scaling-agent-systems-2026-01-28.md

• raw/articles/google-research-science-one-chain-of-evidence-2026-07-30.md

• raw/articles/bnext-discovery-loop-jeff-dean-2026-08-06.md

• raw/articles/google-research-planetary-prediction-engine-earth-ai-2026-08-27.md — PPE 的 geospatial data selection、dataset curation、AutoML 與 prediction loop

• raw/articles/google-research-timesfm-3-multivariate-forecasting-2026-08-31.md — TimesFM-3 的 multivariate forecasting、covariates、quantile output、benchmark 與授權邊界

• raw/articles/sakana-ai-fugu-multi-agent-system-as-model-2026-06-30.md

• raw/articles/aihao-ai-native-engineering-org-2026-06-01.md

• raw/articles/aihao-ai-code-review-bottleneck-2026-04-20.md

• raw/articles/bnext-claude-reflect-ai-usage-reflection-2026-07-15.md — Claude Reflect 使用回顧與 4D AI Fluency，補強個人 AI 協作反思迴圈

• raw/articles/aihao-working-with-ai-compound-2026-05-20.md

• raw/articles/aihao-agents-md-research-and-practices-2026-05-03.md

• raw/articles/aihao-agent-skills-build-and-eval-2026-02-20.md

• raw/articles/aihao-post-mcp-era-skills-vs-mcp-2026-03-12.md

• raw/articles/bnext-chatgpt-skills-tutorial-2026-07-14.md

• raw/articles/bnext-ai-agent-skill-2026-08-13.md

• raw/articles/bnext-third-party-skill-security-cross-platform-2026-08-14.md

• raw/articles/bnext-claude-learning-map-2026-08-19.md — Claude 八段式產品與工作流學習地圖

• raw/articles/bnext-claude-learning-map-revision-2026-08-19.md — Claude 學習地圖修訂版：40 篇教學與 Claude Code /design 畫板預覽

• raw/articles/bnext-claude-code-shortcuts-2026-08-19.md — Claude Code 斜線指令、快捷鍵、context 管理與多 agent 控制面

• raw/articles/bnext-chatgpt-computer-history-2026-08-14.md — Computer History、事件流記憶、context capture 與隱私／prompt injection gate

• raw/articles/bnext-chatgpt-browser-10-automation-tips-2026-08-27.md — 內建瀏覽器、cloud／local／Chrome runtime、browser-use 技巧與權限 gate

• raw/articles/bnext-chatgpt-work-14-features-cloud-vs-local-2026-08-14.md — ChatGPT Work、cloud／local runtime、plugins、排程、語音與 artifact delivery

• raw/articles/bnext-one-person-company-polsia-2026-08-13.md

• raw/articles/bnext-appier-agentic-ai-gross-margin-2026-08-14.md — 企業內部多 agent 協作、人均毛利、token dashboard 與 workflow KPI

• raw/articles/bnext-taiwan-mobile-systex-ai-it-ct-acquisition-2026-08-14.md — IT + CT、AI 資料中心、網路／算力／資安／系統整合與企業 AI 交付鏈

• raw/articles/aihao-pmarca-latent-space-2026-07-01.md

• raw/articles/aihao-agent-experience-designing-for-agents-2026-05-18.md

• raw/articles/bnext-openai-codex-micro-keyboard-2026-07-16.md

• raw/articles/bnext-chatgpt-chrome-extension-codex-guide-2026-07-29.md

• raw/articles/bnext-openai-codex-jason-liu-long-lived-ai-team-2026-07-29.md

• raw/articles/bnext-fubon-insurance-ai-assistants-2026-07-29.md

• raw/articles/aihao-cloud-agent-vs-localhost-agent-2026-05-18.md

• raw/articles/aihao-agent-sandbox-landscape-2026-04-20.md

• raw/articles/aihao-rag-not-vector-search-2026-02-19.md

• raw/articles/aihao-beyond-rag-llamaindex-workshop-2026-07-26.md

• raw/articles/bnext-loop-engineering-ai-coding-loops-2026-06-15.md

• raw/transcripts/kelly-tsai-agent-teams-dynamic-workflows-2026-06-26.md

• raw/transcripts/garychen-claude-codex-auto-review-2026-06-27.md