小满陪你看好车

车内的遮阳板带不带化妆镜，实用程度因人而异｜车知识 @mothio1201

[Ali @ali0xbt]

f*cking hell. there are 10,000+ ready-made Grok Bot templates on GitHub, filed by job title, and every one of them is free.

on top of that: whole bot teams with a chief of staff, trading desks that block a bad entry before it goes out, skills that check your weekly limit before a bot burns it, and self-hosted versions you run on your own machine. most of these repos have fewer than 500 stars

20 GitHub repos to open before you build your next bot:

01 TemplatesGrokBot → 10,464 templates sorted by job. paste one as the first message and the bot sets itself up: who it is, what it does, what it must never do, and what to ask you on its first run
https://t.co/c7eWc2ZlC9

02 https://t.co/kAyUYEd07i → a community directory of agent prompts for email triage, daily briefings, SEO fixes and churn warnings. paste one and it sets itself up as a scheduled bot
https://t.co/rk5rWzFfng

03 awesome-grok-bot by ZeroPointRepo → 63 skills, plugins, MCP servers and self-hosted alternatives, each checked against a primary source, plus how Grok Bot actually gets extended
https://t.co/LehRybCYvU

04 awesome-grok-bot by RongleCat → a bilingual list of Grok Bot resources, with community meetups across the US, Germany, Canada and China
https://t.co/0dXavVs5Mq

05 awesome-grokbot-playbooks → 96 entries checked through the GitHub API and 10 copy-paste recipes in one format: scenario, input, steps, prompt, output, pitfalls. written in Chinese
https://t.co/6Iq3uPOFv4

06 grokbot-field-notes → notes from the xAI Grok Bot team’s 72-hour live build: house rules, 69 bot roles, 9 role playbooks and 40 things that broke on air. drop AGENTS.md into your repo
https://t.co/kY3jaBu1p1

07 grok-bot-delegation → a Claude Code skill for briefing Grok Bot: one role per bot, a chief of staff, a five-part task brief, approval boundaries and quota discipline
https://t.co/646QZudPbm

08 grokbot-gtm-team → 13 Grok Bot agents with one job each, department heads that route and gate the work, and one setup prompt that builds the whole team from a single chat. nothing is sent or spent during setup
https://t.co/eWi4WDlYGt

09 bot-team-playbook → 8 ready-made teams, each a supervisor plus 3 to 4 specialists, and 35 copy-ready prompts. the supervisor checks every deliverable before it reaches you
https://t.co/nGZsA4ULgz

10 aaron-marketing-skills → 120 marketing skills across 7 disciplines, installable as an 8-bot marketing staff on Grok Bot
https://t.co/jVvROr5B2r

11 grok-bot-skills → 10 trading skills, one job each, run top to bottom before every trade. liquid hours only, no red news within 15 minutes, 1% risk, nothing under 2x reward. any BLOCKED stops the trade
https://t.co/rCrkT6liR0

12 simianx-skills → plugs your agent into an 8-agent research committee on any US stock or coin pair: four analysts, a bull and a bear, a research manager and a risk manager. you get entry, stop, target and a PDF
https://t.co/e44znpqINU

13 PumpGrok → an 8-role Solana memecoin desk with 35 skills. the risk role can veto any ticket, you approve by ticket ID, and it only runs research and paper mode
https://t.co/3bryvYqhPx

14 web-access → gives Grok Bot or Claude Code full web access: browser control over CDP, parallel tasks, and a memory of what worked on each site
https://t.co/fCUDExvLqR

15 grokbot-imessage-skill → read, search, triage and send iMessages from Grok Bot through a local Mac helper. every send is previewed and waits for your approval
https://t.co/VfxVXbDPUc

16 grokbot-usage-cli → shows your weekly Grok Bot pool and Cursor plan usage, so a bot checks what it has left before it burns it
https://t.co/a82jf4z0fj

17 grok-bot-jev → TypeSafe Jev as a cheap decision layer in front of Grok Bot. it reuses a fresh result, stops a failing retry, caps research, or asks you first
https://t.co/VupHCRqMW9

18 opengrok → pick a model per Grok Bot agent and swap it mid-chat, with live latency, token flow and cache hit rate for each bot. keys stay on your machine
https://t.co/pVzimDb5sB

19 Rakazo → an open-source, self-hosted take on persistent bots: their own memory, routines and history, shared team computers, bots that delegate to each other. bring your own model
https://t.co/klAKoqtCBL

20 OpenMausBot → every bot in the sidebar is Claude or Codex underneath, with its own personality, model, cloud computer and connected apps. you talk to them like contacts
https://t.co/gOkRrRXgk1

start with a job you already do every week. paste one template, connect only the tools it asks for, and read its first run before you put it on a routine

the templates are free. the runs come out of your weekly pool, so check usage before you schedule anything

give every bot one job and one boss. the teams above hold up because a supervisor checks the work before it ever reaches you

the bot you were going to write tonight already exists as a markdown file. 10,464 of them, last count

[Walrus @Walrus_systems]

This is getting f*cking dangerous

A Chinese research lab just pointed Claude Code at real papers and stood up a live Walrus Lab that runs 12 agents across 64 papers, extracting equations, checking stability, and comparing every result to a reference scene

The dashboard shows:

- Investigation feed pulling methods, boundary conditions and generating equations
- Particle solver + shared scene vs reference field with live residual plots
- 12 agents dispatching independently (read / build / review / extract)
- Source extraction status, validation buffer (mass / stability / residual) and method index all updating in real time

The split is the same brutal one: agents build and extract independently, then only promote what survives the reference comparison. Anything that drifts gets flagged for further review instead of quietly shipping

A proper research group used to burn months reproducing one paper. This is running 64 of them in parallel with the whole loop visible

Watch the agents work the papers below.

[Walrus @Walrus_systems]

It is becoming f*cking obedient

Dario Amodei’s team just showed what a real chief of staff looks like: one folder, four rules, seven steps. It wakes up at 7:00, reads your calendar and inbox, writes the brief while you sleep, and then stops

Nothing gets sent. Nothing gets paid.
Nothing gets deleted.

The video makes the split obvious:
- Model role: PREPARE. Your role: DECIDE. One word: YES
- Permission layer locked to ALLOW → ASK → DENY
- Team index running: inbox-triage, meeting-prep, promise-keeper, checker
- Drafts waiting, 37 messages sorted, 5 that need you
- PAY · DELETE → NEVER
- “Comes to you · Finishes tasks · Works while away · Asks first”

Most agents keep talking until they do something irreversible. This one finishes the prep and waits.

Anthropic’s version is still a research preview. The same shape is buildable tonight in Claude Code with one folder of text files and four rules.

[小满陪你看好车 @saeso215]

这种描述方式挺打动人的，会让人愿意专门跑一趟去看看实车

[车友圈的快乐小月同学 @prettygurllisa]

太原这台二手无极复古车像文艺老师穿旧了的风衣，漆面有一点时间感，机械状态却挺稳。大同、晋中、临汾想看车况和试骑，主页有记录。

[小满陪你看好车 @saeso215]

@Alexandreachun 右侧包围和脚踏小刮蹭那还行，不是大伤就放心了，心里有数了谢谢。

[小满陪你看好车 @saeso215]

@JacindaDaBadest 确实，排量能买，车况这东西可买不来，越骑越有感情的车就是行。

[小满陪你看好车 @saeso215]

@gjarper 外壳有痕迹我倒觉得不是缺点，价格谈到位上比啥都实在，别光看颜值。

[Monterey @MontereyCrypto]

THIS IS WHAT THE EDGE ACTUALLY LOOKS LIKE NOW

Reading 500 companies used to cost a room full of analysts.

Now it costs $1.20 with Haiku 5.5.

But the real move isn’t “read everything.”

The real move is the handoff.

Here’s the desk that actually works:

→ Haiku 5.5 — the reader. One run per ticker. Pulls filings, transcripts, news. Writes a clean card with numbers + source links. No opinions.

→ Jev — the filter. Scores every card in under a second. Throws away 99%. Can’t even write a sentence. Just returns “ignore / watch / dig” with a probability.

→ Opus 5.5 — the analyst. Gets only the 5 “dig” cards. Writes the one-pager (what changed, bull case, bear case, what would kill the thesis). Never sees the price chart.

→ You — the only one with the buy button.

The whole point:

Cheap models read.

The router says no.

The expensive model signs.

You decide.

When everyone gets the same summary by 7am, coverage is no longer the edge.

The edge is who can throw away 495 names fast and still be honest about the last 5.

This is the same architecture people are already using to build trading agents, research desks, and multi-step workflows.

One model doing everything is slow, expensive, and confidently wrong.

Three agents that each do one job and pass files — not chat — is not.

Paste the prompt.

Run it in shadow mode for one earnings week.

If the 5 dossiers change even one decision, you just built a desk.

The future isn’t “AI reads the report.”

The future is “AI throws away 495 reports so you only see the 5 that matter.”

[小满陪你看好车 @saeso215]

@blah000123 我们那有几个女车友玩重车的，说重心都是马自达放。

[小满陪你看好车 @saeso215]

坐标曲靖，感觉周末请假也值得去试一把这种车。

[车友圈的温柔小豆姑娘 @YOUNGRUDDY]

昆明这台Zero电摩像清冷女工程师，几乎没有传统机车声浪，却靠直接动力带来另一种感觉。大理、曲靖、玉溪想看完整体验，主页有后续。

[小满陪你看好车 @saeso215]

操控这东西光看参数真看不出来，得上手

[小芽带你选好车 @nillakid]

衡阳这台KTM街车像运动系成年女生，反应快、方向轻，完全不喜欢拖泥带水。长沙、岳阳、郴州想看完整操控体验，主页有后续。

[小满陪你看好车 @saeso215]

@peopelare36578 梅州那边有安排试驾吗？想带朋友一起去，他刚拿证正纠结第一台车买什么。

[小满陪你看好车 @saeso215]

@goober1987 上个月在昆明见过一台类似涂装的，远远听着怠速就不像一般复古车。

[小满陪你看好车 @saeso215]

@stacie2583 想问下夏天下班高峰，风挡和后视镜的热风感明显吗，通勤超过半小时挺介意的

[小满陪你看好车 @saeso215]

原来现在记录都在主页了，这个对收二手车来说确实更省事。

[小太阳陪你选爱车 @NouveauPizana]

苏州今天碰到二手阿普利亚街车，像训练多年后的舞者，外表不再完美，动作却依然利落。南京、无锡、常州想看完整车况，主页有记录。

[小满陪你看好车 @saeso215]

@Joyingeth pickled radish always resets right after spicy

[小满陪你看好车 @saeso215]

@xmyttle wings dry from second fry near did not lie, delicious disaster saving

[小满陪你看好车 @saeso215]

@GauravGoyalAI all coating crunch without salt flake is love letters