懂车的可爱女孩小杏

选车考虑下车重，太轻高速发飘太重费油｜选车思路 @mothio1201

[懂车的可爱女孩小杏 @cavinderm]

长春这台印第安巡航像穿长裙的轻熟女，车身修长，节奏慢，骑久以后特别容易沉进去。吉林、四平、松原想看更多，主页继续。

[sunick @Serantych]

stop giving GrokBot more questions
give it something to take off your plate

100 tasks you can delegate:

→ research
→ content
→ operations
→ analysis
→ communication
→ coding
→ monitoring
→ support
→ planning
→ creative

plus a prompt formula, automation ideas and approval checkpoints

100 tasks
10 categories
one sheet

save this before you do the next task yourself ↓

[Mojunel.eth @mojunel]

Paste once. Basically a five-check read-only phone audit.

Claude Code in read-only mode. Own device. Nothing written.

The prompt drops. Five passes fire in sequence:

→ WALK starts with every installed app, hidden ones included. One case out there: an app disguised as "System Service," no icon, no store listing, location pinged every 2 minutes for 11 weeks, per a posted walkthrough
→ EXPOSE pulls configuration profiles on iPhone, device-admin and accessibility on Android. Who installed each, what it controls
→ CATCH flags any app holding always-on location, mic or camera it does not need
→ HUNT maps anything relaunching on boot or connecting to raw IPs on a fixed schedule
→ DROP. One verdict line: clean or compromised. Three sorted lists: Safe. Worth removing. Needs a professional

Each flag carries severity: low, medium or high. What it reaches, how long it has been active, where it sends data.

Skip the boundary and you burn the evidence. The five passes work because nothing writes to the phone. What the session reads depends on the access you provide.

If something hits the last list: screenshot first, passwords from a different device, professional before deletion.

The terminal below walks the five passes to the verdict ↓

[懂车的可爱女孩小杏 @cavinderm]

@0xbobaaa cold mornings equals wire bite in that side route at least fast

[懂车的可爱女孩小杏 @cavinderm]

@MAXdeg0 my touring jacket now somehow smells like straight storm not jokes

[懂车的可爱女孩小杏 @cavinderm]

@MAXdeg0 cold mornings equals wire bite in that side route at least fast

[Kaiju @kaijucode]

THIS IS AN ABSOLUTE GOLDMINE FOR REVERSE ENGINEERING

people are decompiling 30-year-old games with Claude, and almost nobody checks if the output matches the binary.

The idea: three Claude models share one set of evidence files, and the original binary decides who’s correct. No address, xref or test log means the claim stays UNKNOWN.

You start with a tiny C file where you already know every answer:

cc -O2 -o samples/record samples/record.c

here’s who does what:
→ Sonnet 5.5 ($2/$10) maps the binary into function_map.md
→ Opus 5.5 ($4/$20) rebuilds the tricky functions with 2 competing readings
→ the original binary runs the tests and logs them to evidence/
→ Fable 5.1 ($10/$50) only reviews a theory that already passed
the trap this catches:
→ Claude sees the checksum loop and calls it CRC32
→ the constants 2166136261 and 16777619 say FNV-1a
→ only a run against the binary proves which one
100K in + 10K out costs about $0.30 on Sonnet, $0.60 on Opus and $1.50 on Fable. Most people burn their Fable money on stuff Sonnet handles fine.
the full tree
CLAUDE.md sets the rules (OBSERVED / INFERRED / UNKNOWN)
Sonnet 5.5 maps the binary
Opus 5.5 writes the spec, 2 readings and minimum tests
the original binary runs the tests
Fable 5.1 reviews only if both readings survive
Sonnet 5.5 turns the proven spec into code
paste the tree and this prompt into Claude Code ↓
“Set up a reverse engineering workflow from this tree:
Create CLAUDE.md. Every claim follows CLAIM | EVIDENCE | CONFIDENCE | HOW TO VERIFY and is tagged OBSERVED, INFERRED or UNKNOWN.
On Sonnet 5.5, inventory [BINARY] into notes/function_map.md with ADDRESS | NAME | PROPOSED_ROLE | EVIDENCE | CONFIDENCE.
On Opus 5.5, reconstruct [FUNCTION] into notes/spec.md with two interpretations and the minimum tests that separate them.
Run the tests against the original binary and log every result in evidence/.
If both readings survive, ask Fable 5.1 for the 3 highest-impact ways the spec could be wrong.
Show me the function map first and wait for my go.”

bookmark this, you’ll need it later.

[Slonski @Slonski_rt]

THIS IS F*CKING GOLD: 100 CLAUDE SKILLS ON 1 SCREEN
767K+ GITHUB STARS. ALL FREE. SAVE YOUR AI COURSE MONEY

a cheat sheet for everything you can teach Claude to do

10 categories covering collections, memory, workflow, marketing, design, data, security, docs, dev and automation

the skills worth grabbing first:

→ everything-claude-code, 219k stars, a giant pack of ready skills
→ ui-ux-pro-max-skill, 95k, for cleaner design
→ claude-mem, 84k, gives Claude a memory between chats
→ superpowers by obra, 42k, a full workflow for coding
→ obsidian-skills, 36k, connects Claude to your notes
→ marketingskills, 35k, for marketing work
→ planning-with-files, 24k, keeps plans in files so Claude does not lose track

the useful part:

only 14 of the 100 are official from Anthropic: docx, pdf, pptx, xlsx, skill-creator and more
the other 86 come from the community

start with the official ones for daily work
use community packs to explore what else Claude can do

the workflow:

pick a skill → read its SKILL.md → install it → give Claude a real task → keep it or delete it

the 3 i would try first:

→ skill-creator, so Claude builds new skills for you
→ claude-mem, so it stops forgetting you
→ superpowers, if you write code

save this, then pick 3 and test them this week

[懂车的可爱女孩小杏 @cavinderm]

@RabitManH 市区红绿灯多其实还好，趴不趴主要看你平时骑姿习惯，我个子不高感觉还能接受，要不你先去坐一下实车？

[懂车的可爱女孩小杏 @cavinderm]

莆田这台杜卡迪Scrambler Nightshift很会拿捏气氛，油箱线条漂亮，低把也有味道。福州、泉州、厦门想看完整试驾，主页继续。

[懂车的可爱女孩小杏 @cavinderm]

@dimplesce 对，低把坐姿更往前压，脖子和手腕都得慢慢适应，你泉州骑老款是跑长途还是市区多？

[懂车的可爱女孩小杏 @cavinderm]

@l0lsmileyface14 哈哈福州确实有几段路拍出来车特别脏，你那边常跑哪个片点？

[懂车的可爱女孩小杏 @cavinderm]

@kelwil4 太对了，低把一坐上去整个人就前倾了，短途帅长途腰会记账。你要真跑长途的话我可以顺便说说坐垫那点事。

[懂车的可爱女孩小杏 @cavinderm]

@Melissabby32 泉州这两天我在对时间，过两天把具体日期发主页，你大概哪个周末方便？我按人凑一凑。

[懂车的可爱女孩小杏 @cavinderm]

@Joyfullchaos 厦门海边那个场景太对了，折线就是得斜阳才出味。你那天看到的是哑光那台还是亮面的？

[懂车的可爱女孩小杏 @cavinderm]

@Perkinsracing41 低把是原厂Nightshift就这么配的，没动过。福州试驾我应该会安排，你之前骑的是什么车？

[懂车的可爱女孩小杏 @cavinderm]

@potpot18 配色这块确实是图片骗人，我这台实车比拍照暗一点、但有质感，你在哪个城市？要能碰面试驾你自己看一眼最准。

[Mojunel.eth @mojunel]

agent code sat on bare metal long enough

Perplexity open-sourced SPACE Oct 15. Millions of sandbox runs daily per Perplexity.

10 open-source GitHub repos that jail your agent's code:

96,151 stars, Oct 18. 9 Apache-2.0. All pushed since Oct 14.

JAIL THE PROCESS

01. sandbox-runtime ★5,577: https://t.co/zSI3arkjA6 → OS-level process limits
02. bubblewrap ★9,104: https://t.co/sICuAxgZdL → unprivileged namespace sandbox
03. gvisor ★19,702: https://t.co/h8MeMb2OUX → user-space kernel isolation

BOOT A DISPOSABLE VM

04. firecracker ★37,461: https://t.co/5ceaps6QlG → lightweight KVM microVMs
05. kata-containers ★8,960: https://t.co/kXiA15eZ8C → VM-isolated container runtime
06. libkrun ★2,794: https://t.co/aBgT3enVxz → embeddable in-process VM
07. smolvm ★6,759: https://t.co/8BcBuXA2K2 → branchable microVMs, sub-second (README)

SERVE SANDBOXES TO AGENTS

08. SPACE ★1,470: https://t.co/xE5roziPGa → Firecracker sandbox HTTP API
09. infra ★1,739: https://t.co/J2XHcjSWLT → E2B self-hosted runtime
10. boxlite ★2,585: https://t.co/zYK7v7DDW8 → five-language micro-VM SDK

the stack:
code → jail → kernel → microVM → API → fork → resume

egress and credentials need their own layer

not the model. the jail under it ⭣

[懂车的可爱女孩小杏 @cavinderm]

@Gyome1_ expired in 2015 gave colors that single every print the editing

[懂车的可爱女孩小杏 @cavinderm]

@RuOnChain1 half my editing is arguing with the grain mix proudly