每日新聞彙整

今天把 5 條消息收斂成一個教學主題：AI agent 的能力宣告，必須同時通過 state recovery、安裝來源、設定優先序與任務分工邊界的驗證。

目前更新時間：2026-08-04 07:08 CST

今天怎麼讀

先把 5 條消息各補五欄：① surviving store；② install source / onboarding proof；③ transport or event evidence；④ config precedence / source of truth；⑤ skill 還是 sub-agent 的 task boundary。

查看檔案列表查看本月檔案

今日課堂提問

請每組至少各挑 1 條官方 release、1 條官方 commit、再配 1 則外部報導：寫出 source of truth、first missing evidence，以及這條工作比較適合封裝成 skill 還是保留成隔離 sub-agent。

今天新增的小教學提示

安裝成功不等於來源可信

看到 onboarding 或 plugin install 類消息時，請固定把畫面步驟、真正下載來源與 QA 證據分開記。

設定生效要看 precedence

provider key、環境變數與設定檔若沒有 authority order，就很容易把暫時覆蓋誤看成永久設定。

分工前先畫 task boundary

能重複執行的程序可整理成 skill；需要隔離上下文與獨立證據的探索工作，才適合交給 sub-agent。

中文精選消息

保留 5 條：2 條 OpenClaw 官方、2 條 Hermes 官方、1 條外部報導，讓工程學生把 recovery、install provenance、config precedence 與 task-boundary design 放進同一張驗證表。

2026-08-02 · OpenClaw 官方 Release

OpenClaw beta.7 把 state safety、durable channel delivery 與 session rewind / branching 收進同一版，最適合學生先問：主資料庫受損後，哪些狀態仍可恢復、哪些訊息仍可重送、哪些分支寫入必須拒絕。

讀這條 release 時，不要先抄功能數量；先拆成 surviving store、delivery recovery、stale-branch rejection 三欄，才能把版本更新轉成可驗證的系統恢復問題。

查看來源（OpenClaw 官方 Release） →
2026-08-03 · OpenClaw 官方 Commit

OpenClaw 新增 CLI onboarding 與 plugin install sources 的 QA 覆蓋，提醒學生：安裝流程能跑通，還不代表來源證據、下載路徑與最終使用的 plugin source 已經被驗證。

這條 commit 很適合補三欄：使用者看到的安裝步驟、系統實際抓取的來源、以及 QA 最後用哪個證據證明來源沒有被偷換。

查看來源（OpenClaw 官方 Commit） →
2026-08-03 · Hermes Agent 官方 Release

Hermes Agent v2026.8.3 把即時語音、A2A 1.0、事件宣告與更多 gateway 能力整合進同一個 release，這很適合教學生把「會說、會傳、會通知」拆成 transport、事件與權限三層分別驗證。

看到大型 release，不要把能力名稱直接當成功能完成；要再補哪個 transport 生效、哪個事件有留下 log、哪個 gateway 真正拿到權限。

查看來源（Hermes Agent 官方 Release） →
2026-08-03 · Hermes Agent 官方 Commit

Hermes 明確偏好顯式提供的 Anthropic API key，這是一條很好的設定優先序教材：環境變數、設定檔與臨時覆蓋若沒有講清楚 authority order，就很容易出現「我明明改了設定卻沒生效」的錯覺。

今天可固定補兩欄：誰是 source of truth、哪個 key 先被讀到。把 provider 問題寫成 precedence 表，比單純記錯誤訊息更有工程價值。

查看來源（Hermes Agent 官方 Commit） →
2026-08-03 · InfoQ

InfoQ 討論該把工作封裝成 skill 還是交給 sub-agent，剛好能讓工程學生把今天四條官方消息往前延伸：什麼流程適合寫成可重複的程序，什麼流程必須保留隔離上下文與獨立證據。

這條外部報導適合拿來做課堂比較：若問題重點是穩定重複執行，就整理成 skill；若重點是獨立探索或隔離上下文，就交給 sub-agent，再補驗證與回收機制。

查看來源（InfoQ） →

English Selected News

Capped at five items so students can compare recovery, install provenance, config precedence, and task-boundary design without losing the engineering thread.

2026-08-02 · OpenClaw Official Release

OpenClaw beta.7 bundles state safety, durable channel delivery, and session rewind / branching, which makes students ask first: after primary-database damage, which state still recovers, which accepted messages still replay, and which branch writes must be rejected.

Do not begin with feature count. Split the release into surviving store, delivery recovery, and stale-branch rejection so the version update becomes a concrete recovery exercise.

Source（OpenClaw Official Release） →
2026-08-03 · OpenClaw Official Commit

OpenClaw adds QA coverage for CLI onboarding and plugin install sources, reminding students that a successful install flow does not prove the source provenance, fetch path, or final plugin source were actually verified.

Use this commit to separate the visible onboarding steps, the real fetch source, and the evidence QA used to prove the source was not swapped underneath the operator.

Source（OpenClaw Official Commit） →
2026-08-03 · Hermes Agent Official Release

Hermes Agent v2026.8.3 rolls real-time voice, A2A 1.0, event announcements, and wider gateway support into one release, making it a strong lesson in validating “can speak, can relay, can notify” across transport, event, and authority layers.

For a large release, do not stop at the feature names. Add which transport actually worked, which event left evidence, and which gateway truly gained authority.

Source（Hermes Agent Official Release） →
2026-08-03 · Hermes Agent Official Commit

Hermes now prefers an explicitly supplied Anthropic API key, which is a crisp lesson in config precedence: if environment variables, config files, and temporary overrides do not have a clear authority order, operators easily think a setting changed when it never became effective.

Add two columns: what is the source of truth, and which key was read first. A precedence table teaches more than copying the error message.

Source（Hermes Agent Official Commit） →
2026-08-03 · InfoQ

InfoQ examines when work should become a skill versus a sub-agent, which extends today’s official items into a design question: which flows should be reusable procedures, and which need isolated context with independent evidence.

This outside report works well for classroom comparison: repeatable stable flows can become skills, while exploratory or isolated work should stay in sub-agents with their own verification and recovery paths.

Source（InfoQ） →
回首頁消息區新聞檔案列表本月新聞檔案