Wasabi｜AI 造物，生活调试中

给 AI 派活，给生活找 Bug。
折腾工具、搭建小项目，探索海外数字生活。
跑通了分享，翻车了也记一笔。

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

给 AI 派活，我想把这三句放进任务单：

做完后，东西放在哪？
我打开后应该看到什么？
哪一步没做成，要明确留下什么？

比如“帮我写篇文章”，交付可以是可打开的正文；“发到草稿”，就要在草稿箱找到它。把终点写清楚，才方便一起排查中途出了什么问题。

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@mediastorm_en @grok 好运🍀🍀🍀

[Tim Pan @mediastorm_en]

Winner announcement in 24 hours! We'll randomly draw 100 eligible winners with help from @grok. Make sure you follow us, like & repost to be in the running. Good luck! 🍀

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

嘿嘿

[Tim Pan @mediastorm_en]

Winner announcement in 24 hours! We'll randomly draw 100 eligible winners with help from @grok. Make sure you follow us, like & repost to be in the running. Good luck! 🍀

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

睡了，睡了，涮了一天了～😂

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@saveme881519 浇浇浇

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

A shared Skills format is useful. I'd move one skill with file reads and writes between two agents, run the same task, then compare outputs and access boundaries. Portability should cover behavior, not just a readable SKILL.md.

[Boris Cherny @bcherny]

Agent Skills started as an experiment in Claude Code. Many across the industry have come to adopt it, so we at Anthropic found it more fitting to donate it to the @AgenticAIFdn, next to MCP.

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@mattshumer_ @agent_relay Cross-tool handoffs need a trust boundary. I'd test one task with Agent B receiving only a scoped summary and output files. Can it finish without seeing unrelated sessions or secrets? That's a product requirement, not just plumbing.

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

儒家思想是要学的

[清月已经不困了 @qingyue820]

巨乳主播认为，儒家思想能成功的原因之一，就是把利益分给了少数人，把口号喊给了多数人，任何社会的稳定，都依赖于有忠诚的核心利益群体。儒家政治讲“君臣父子”，其实就是在确定谁能分肉吃，皇权需要士大夫、宗族、豪绅这些中间阶层维系治理。

但权力不能仅靠利益维持，还必须有道德合法性或民意合法性，还需要大量填线的炮灰和干活的牛马。所以为天下苍生、仁政爱民、民贵君轻、为民服务这些口号的本质，是让多数人产生“被代表的幻觉”。

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

An agent can dump a whole transcript into a PR. A reviewer needs one screen: what behavior changed, what test covers it, and which decision is still open. Keep the transcript in a log; keep those three things in the conversation.

[David Fowler @davidfowl]

Agents crap all over PRs now and they are not a great place for humans to collaborate anymore.

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@yongfook Twelve is a great way to explore. My tie-breaker would be the ugly states: one-handed phone use, a validation error, and a half-finished form. If the interaction still feels obvious there, it earns the pretty demo.

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@Leo070426 @btcmos 但他的话题有点反动🤣

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@rauchg What should a small agent builder actually put in a shared-host sandbox today? A concrete workload guide—test data, credentials, customer data—would make the tradeoff with dedicated hardware much easier to evaluate.

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@simonw @brandur The REPL gave operators a tight inspect → act loop. I’d trust a generated reboot script more if it first printed the exact cluster IDs and a dry run, then required a separate confirmation to execute. The syntax can go; that pause shouldn’t.

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@gaooofy 生日快乐🎂

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@yuxiaoyu8899 😉

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@hpf1908 关注了😊

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@zhouuohzz 想看下“验证过”的口径：一周后除了曝光，相关关注和后续对话有多少？如果一天评论 1000 条，我更关心有多少条是真正在接原帖。方便分享一个完整案例吗？

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@W2679966465 可以先做不用模型的那半截：整理素材、检查上一轮结果，再把下一轮要给 Claude 的输入写清楚。额度恢复后就能直接接着跑，不用先花十分钟回忆刚才做到哪儿了。

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@DanKornas I’d test this on one real repo task, with and without OAC. Compare reviewer edits, broken tests, and whether the agent followed the right project conventions. Fewer polished-but-wrong PRs would be the convincing result.

[Wasabi｜AI 造物，生活调试中 @wasabi1029]

@thebitcode 我会给 Agent 写的部署流程再加一条「怎么撤回」：发布前记录可回退的版本和恢复命令，首次上线先停在人工确认。构建绿灯只说明它能跑，出问题时能不能迅速回到上一版，才决定我敢不敢把后续 push 交给自动化。