the model isn't the problem. your setup is | ai skills, agents, prompts
Washington
Joined September 2026
- Tweets919
- Following44
- Followers106
- Likes1.2K
Mia Khalifa was one of the most searched names on the biggest adult site on earth.
Over a billion views on her videos.
Her cut: $12,000.
She was 21. Three months. About 12 scenes at roughly $1,000 each. She quit in early 2015.
The videos never stopped. Millions in revenue for the companies. Her share of ongoing profits: zero.
She doesn’t own the videos, the photos, or even the domain with her own name on it. Her team tried to buy the rights back. No deal.
The studio later said she was paid $178,000 and filmed more scenes. She says most of that was for a separate social-media job.
Two versions. They can’t both be true. Either way, the clips are still running.
In her words: “It makes me feel like I lost all right to my privacy.”
She signed the contract.
Does the internet own her face forever - or should anyone be able to take it back?
Sophie Rain has shown ~$100M+ in OnlyFans gross. High-profile people with real money online are constant targets.
A recent case shows how little it takes: a user lost thousands after three days on open hotel Wi-Fi. No phishing link. No leaked seed phrase. Just a fake approval prompt that looked normal while he was on a legitimate site.
Here’s the usual play:
→ Attacker joins the same local network (hotel, Airbnb, café, or a compromised router).
→ Injects or sits in the middle of traffic.
→ You open your wallet or exchange as usual.
→ A “reconnect,” “verify,” or approval request appears.
→ You sign once. Funds move later (or immediately).
On-chain there is no undo.
How to stay uncooked:
→ Never touch a wallet or exchange on public/shared Wi-Fi. Use a hotspot or VPN.
→ Keep serious size on a hardware wallet.
→ Read every signature on the device before confirming—especially permissions and unlimited approvals.
→ Lock down your own router (change defaults, disable remote access).
Your bag is only as safe as the network you open it on.
Save this before the next trip.
Sophie Rain has shown ~$100M+ in OnlyFans gross. High-profile people with real money online are constant targets.
A recent case shows how little it takes: a user lost thousands after three days on open hotel Wi-Fi. No phishing link. No leaked seed phrase. Just a fake approval prompt that looked normal while he was on a legitimate site.
Here’s the usual play:
→ Attacker joins the same local network (hotel, Airbnb, café, or a compromised router).
→ Injects or sits in the middle of traffic.
→ You open your wallet or exchange as usual.
→ A “reconnect,” “verify,” or approval request appears.
→ You sign once. Funds move later (or immediately).
On-chain there is no undo.
How to stay uncooked:
→ Never touch a wallet or exchange on public/shared Wi-Fi. Use a hotspot or VPN.
→ Keep serious size on a hardware wallet.
→ Read every signature on the device before confirming—especially permissions and unlimited approvals.
→ Lock down your own router (change defaults, disable remote access).
Your bag is only as safe as the network you open it on.
Save this before the next trip.
THIS CABLE CHARGES YOUR PHONE. BUT PLUG IT INTO YOUR LAPTOP AND IT COULD START TYPING COMMANDS WITHOUT YOU.
same plastic. same logo. charges like any other cable. you’d never look at it twice.
inside there’s a chip smaller than a fingernail. plug it in and your laptop thinks a keyboard just connected. then it types on its own. opens a terminal, runs commands, closes it before you look up.
some versions go further and log your keystrokes. passwords included.
this isn’t sci-fi. it’s called the O.МG cable, and it’s sold to security testers right now.
will someone target you with one? honestly, probably not.
but the trick behind it is everywhere:
→ your laptop trusts anything that says “i’m a keyboard”
→ public chargers and “free” cables are a blind spot
→ nobody checks what a cable actually is
simple habits:
→ use your own cable. always
→ “trust this computer?” at a public charger = tap no
→ a cheap usb data blocker for public ports
borrowed a cable or found one in the office? i made a prompt that has claude code check it in 2 minutes.
it’s in the first reply.
YOUR WIFI CAN DETECT MOVEMENT WITHOUT A CAMERA. I ASKED CLAUDE CODE TO TURN THAT IDEA INTO A 3D RADAR.
And the result is pretty damn wild.
A week ago, I turned my Wi-Fi router into a radar using Claude Code. No camera needed.
Claude rewrote the router’s firmware in just 10 minutes, then built code that analyzes how Wi-Fi signals bend around human bodies, turning them into moving silhouettes on a 3D map of my apartment.
I’d been running it every night, just to see what it could detect. But last night, I couldn’t sleep. So I opened the radar map, and something caught my attention.
Three human silhouettes. Tracking people as they move through rooms. And even the ability to monitor a person's heart rate
I was scared by that third silhouette because I've been living with my girlfriend for a year now. I asked Claude about it, but he didn't say anything.
This technology has both helped and frightened people at the same time:
WiFi CSI measures how wireless signals change as people move through a space.
The basic motion detection logic:
csi = np.abs(wifi.get_csi())
motion = np.mean(np.abs(csi - baseline))
if motion > threshold:
radar.flag_activity()
Claude Code helped me build a 3D visualization around that concept.
→ CSI: 64 real-time motion signals.
→ Three.js: lifelike human figures and a 3D apartment.
→ Tracking: real steps, movement between rooms, and persistent identifiers.
→ Vital signs: breathing and heart rate readings.
And here's what's actually insane.
Researchers are already using real WiFi CSI to detect movement, estimate breathing, and explore contactless heart-rate monitoring.
My video shows just how insanely fast this tech is evolving. We're talking days, not months.
Today Claude Code builds a WiFi radar. Tomorrow it'll probably make my damn toaster cook breakfast at 9 AM lol.
Soon, your WiFi won't just connect your devices. It could actually sense what's happening inside your home.
And that third figure on my radar? Still have no clue who it was supposed to be...
So what do you think? Pretty sick innovation or a straight-up privacy nightmare?
11 PEDESTRIANS. 6 VEHICLES. 1 TAXI. ONE CAMERA TRACKING EVERYTHING.
No one watching the monitor. No security team manually tracking every movement.
Just Claude Code turning ordinary street footage into a damn impressive AI traffic monitoring system.
And honestly, it handled the job better than I expected.
Watch the screen.
A pedestrian enters the crosswalk. Detected, assigned an ID, tracked.
A truck moves through the intersection. Classified.
A taxi appears among regular cars. Identified separately.
Every object gets tracked. Every movement generates data. And the dashboard updates as the scene changes.
33 objects tracked. 0 alerts. All from one camera.
Here's where Claude Code really impressed me.
Instead of spending days manually connecting detection models, tracking algorithms, and event logging, I used Claude Code to help bring the entire pipeline together.
The detection stage uses the same basic idea as this simplified YOLOX snippet:
frame = letterbox(frame, 640)
out = yoloх.run(None, {"images": frame})[0]
people = out[out[:, 4] * out[:, 5] > 0.12]
tracks = tracker.update(people)
Then the system connects several technologies:
→ OpenCV processes the video frames.
→ YOLO / YOLOX detects pedestrians and vehicles.
→ ByteTrack maintains tracking IDs across frames.
→ Python handles crosswalk zones, event logs, and alerts.
And that's the real achievement.
Claude Code didn't invent computer vision. It helped turn several complex technologies into one working system.
A regular CCTV camera records traffic.
This system converts that footage into structured data you can actually use.
And this is only one of my experiments.
I tested 7 ways Claude Code can interact with the real world, from CCTV monitoring to device diagnostics and security automation.
The complete breakdown, code examples, and experiments are in my article below.
AI coding agents aren't just writing code anymore. They're helping us build tools that interact with the world around us.
If you liked this, read my article. Lots more info like this there ↓
A $5 CABLE + CLAUDE CODE ( FREE VERSION ). YOUR CAR JUST GOT HACKED. NO CAP.
This guy plugged a cheap OBD2 adapter into his dad's car and gave Claude Code one job:
"Figure out how to talk to this car and build me a live dashboard."
5 minutes later, things got pretty damn interesting.
Real engine data. Live temperatures. Diagnostic readings. All flowing into an app Claude helped build from scratch.
Here's what was actually happening under the hood.
Claude wasn't just drawing some fancy dashboard. It had to figure out how to communicate with a real car.
And the tech behind it is surprisingly simple:
→ ELM327 + OBD-II: A cheap adapter connects the laptop to the car's diagnostic interface, giving access to supported ECU data.
→ SAE J1979: Commands like 010C request engine RPM, 0105 reads coolant temperature, and 010D checks vehicle speed.
→ Python + python-OBD: Sends diagnostic queries, reads sensor values and converts raw responses into actual measurements.
→ pySerial: Handles communication with compatible adapters, including connection errors and timeouts.
→ Claude Code: Helps write the connection logic, fix broken scripts and turn incoming vehicle data into a working application.
→ HTML + CSS + JavaScript: Can transform those numbers into live gauges, temperature indicators and a custom dashboard.
→ DTC diagnostics: Reads supported fault codes so you can understand what that annoying check-engine light is complaining about.
And here's the funny part.
You don't need 500 lines of code just to ask your engine what it's doing.
A basic read-only query looks like this:
```python
import obd
car = obd.OBD()
rpm = car.query(obd.commands.RPM)
temp = car.query(obd.commands.COOLANT_TEMP)
print(rpm.value, temp.value)
```
A few lines of Python. Real engine data.
The dashboard shows 69°C coolant temperature, battery voltage and diagnostic status.
And the cool part? Claude can customize everything.
Racing-style RPM gauges? Done.
Engine fault explanations? Easy.
A full Knight Rider-style dashboard? Hell yeah, lol.
But here's the twist.
He didn't bypass the car's security. He made it communicate with custom AI-built software.
5 minutes. One prompt. A $5 cable.
Next time your check-engine light comes on, imagine opening your own diagnostic app instead of guessing what's wrong.
Authored by @Vectorxeth
A ROBBERY WAS HAPPENING. NOBODY WAS WATCHING THE CAMERAS. CLAUDE CODE CHANGED THAT.
One CCTV camera. Python. YOLOv4-tiny. ByteTrack.
I gave Claude Code one task:
"Detect threats, track suspects, and trigger an emergency response."
Here's how the system works.
1. YOLOv4-tiny: Detection
The first challenge was teaching the camera to actually see what's happening.
YOLOv4-tiny processes CCTV footage frame by frame, detecting people and drawing bounding boxes around them.
Each detection comes with coordinates and a confidence score, helping filter out false positives.
And the best part? It's lightweight enough to run on a regular CPU without an expensive GPU.
detections = yolo.detect(frame)
people = [d for d in detections if d.class_id == 0]
2. ByteTrack: Tracking
Keeps consistent IDs as people move through the store.
tracks = bytetrack.update(people)
for person in tracks:
update_position(рersоn.іd, person.bbox)
PERSON_01 → RESTRICTED AREA
PERSON_02 → NEAR CASHIER
PERSON_03 → RUNNING
3. ThreatAnalyzer: Connecting the dots
One suspicious movement means little. Multiple warning signs? That's where shit gets serious.
risk = analyzer.score(
tracks=tracks,
zones=restricted_zones,
events=detected_events
)
if risk >= 0.85:
trigger_alert("CRITICAL")
4. Emergency Response: The crazy part
Watch what happens when the robbery escalates.
The cashier is threatened. Someone forces a rear door.
The system connects the events and flags a critical threat.
And instead of just recording everything, it kicks off an emergency workflow.
if threat.level == "CRITICAL":
save_cctv_clip()
notify_owner()
generate_evidence_report()
initiate_emergency_workflow()
→ CCTV footage automatically preserved
→ Suspects tracked with persistent IDs
→ Evidence and timestamps collected
→ Security alerts triggered
→ Simulated 911 call initiated
The video even shows a conversation with 911, conducted by AI, during which details of the incident are relayed to the dispatcher.
There is no person behind the monitor. No one is manually reviewing hours of video footage
The code snippets illustrate the system's architecture.
But imagine this deployed with proper testing and human oversight.
Someone threatens your cashier at 3 AM.
Claude's software detects the danger and starts preparing the response while the crime is still happening.
That's the kind of automation that could save precious seconds.
A normal camera records a robbery.
An AI-powered system could help someone react before it's over.
Would you trust this system to protect your store overnight?
HOLY SH*T. CLAUDE CODE TURNED A BASIC CCTV CAMERA INTO A ROBBERY EVIDENCE SYSTEM.
The camera already caught the thieves.
Claude Code made the footage actually useful.
I gave it one job:
“Track everyone involved. Find the clearest frames of each suspect. Save the evidence.”
Then it started working.
[0:01] Multiple people rush from the store.
Every person gets a persistent ID.
PERSON_01
PERSON_02
PERSON_03
Even when someone disappears behind another person and comes back, the tracker tries to keep the same ID.
But this is where it got stupidly interesting.
Normal dev: “ok, can’t be done.”
Claude Code: “bet.”
Instead of giving up because the full stack wasn’t available, it rebuilt the YOLOv4-tiny forward pass in NumPy.
Every conv layer boiled down to something like:
win = sliding_window_view(x, (k, k), axis=(0, 1))[::stride, ::stride]
cols = win.transpose(0, 1, 3, 4, 2).reshape(oh * ow, k * k * C)
z = cols @ W + b
y = np.where(z > 0, z, 0.1 * z)
0.27 seconds per frame.
On a CPU.
No GPU.
No cloud.
No expensive security box.
Built different.
Then Claude Code goes through the tracked suspects and searches for the clearest usable frame of each face.
Blurry? Skip it.
Face covered? Skip it.
Better angle later?
Replace the old image.
By the end, each suspect gets an evidence folder:
PERSON_01/
face_best.jpg
full_body.jpg
first_seen.jpg
last_seen.jpg
movement.json
Then it packages:
→ suspect screenshots
→ timestamps
→ movement history
→ original clips
→ incident summary
into one report ready for security or police review.
That’s the part people miss.
The impressive thing isn’t the red boxes.
It’s taking chaotic CCTV footage and turning it into organized evidence automatically.
And no, it’s not magically discovering someone’s real identity from their face.
It’s doing something more grounded:
tracking the same person, finding the strongest visual evidence, and preparing it for a human to act on.
A few years ago we were impressed when AI could write a decent Python function.
Now Claude Code can rebuild missing computer vision pieces, process CCTV on a CPU, track suspects, extract faces, and assemble an incident package.
The camera always had the footage.
Claude Code gave it a brain.
HOLY SH*T. CLAUDE CODE TURNED A BASIC CCTV CAMERA INTO A ROBBERY EVIDENCE SYSTEM.
The camera already caught the thieves.
Claude Code made the footage actually useful.
I gave it one job:
“Track everyone involved. Find the clearest frames of each suspect. Save the evidence.”
Then it started working.
[0:01] Multiple people rush from the store.
Every person gets a persistent ID.
PERSON_01
PERSON_02
PERSON_03
Even when someone disappears behind another person and comes back, the tracker tries to keep the same ID.
But this is where it got stupidly interesting.
Normal dev: “ok, can’t be done.”
Claude Code: “bet.”
Instead of giving up because the full stack wasn’t available, it rebuilt the YOLOv4-tiny forward pass in NumPy.
Every conv layer boiled down to something like:
win = sliding_window_view(x, (k, k), axis=(0, 1))[::stride, ::stride]
cols = win.transpose(0, 1, 3, 4, 2).reshape(oh * ow, k * k * C)
z = cols @ W + b
y = np.where(z > 0, z, 0.1 * z)
0.27 seconds per frame.
On a CPU.
No GPU.
No cloud.
No expensive security box.
Built different.
Then Claude Code goes through the tracked suspects and searches for the clearest usable frame of each face.
Blurry? Skip it.
Face covered? Skip it.
Better angle later?
Replace the old image.
By the end, each suspect gets an evidence folder:
PERSON_01/
face_best.jpg
full_body.jpg
first_seen.jpg
last_seen.jpg
movement.json
Then it packages:
→ suspect screenshots
→ timestamps
→ movement history
→ original clips
→ incident summary
into one report ready for security or police review.
That’s the part people miss.
The impressive thing isn’t the red boxes.
It’s taking chaotic CCTV footage and turning it into organized evidence automatically.
And no, it’s not magically discovering someone’s real identity from their face.
It’s doing something more grounded:
tracking the same person, finding the strongest visual evidence, and preparing it for a human to act on.
A few years ago we were impressed when AI could write a decent Python function.
Now Claude Code can rebuild missing computer vision pieces, process CCTV on a CPU, track suspects, extract faces, and assemble an incident package.
The camera always had the footage.
Claude Code gave it a brain.
£25,000 BMW WAS STOLEN IN UNDER A MINUTE. THE SECURITY CAMERA CAUGHT EVERYTHING.
Holy sh*t. Two guys, one keyless relay box, 40 seconds. No smashed glass. No keys. The owner found out the next morning.
So I gave the footage to Claude Code with one job:
“Watch the driveway. Track everyone. Tell me the second it turns into a theft.”
No GPU. No cloud. No $5k security system. One laptop CPU.
Watch the screen:
[0:02] Two blue boxes. PERSON_01, PERSON_02.
[0:04] PERSON_01 pulls out a handheld device. Status flips to SUSPICIOUS.
[0:11] The car unlocks. No owner anywhere near it.
→ Both boxes turn red. PERSON → THIEF.
[0:35] Engine starts. VEHICLE THEFT DETECTED. Footage saved.
Here’s what Claude Code actually wrote.
Step 1: find every person in every frame.
frame = letterbox(caр.reаd(), 640)
out = yoloх.run(None, {"images": frame})[0]
people = out[out[:, 4] * out[:, 5] > 0.12] # class 0 = person
boxes = nms(people, iou=0.45)
1,228 frames. 2,214 detections. Night footage, motion blur, hoodies.
Step 2: don’t lose them. This is where most trackers break. A guy walks out of frame, comes back, and gets a new ID like he’s a stranger. Claude Code added this:
# he left the frame. don't give him a new ID
same_guy = (
gap_s < 4
and dist(last_seen, new_pos) < 260
and new_h > 0.45 * last_h # crouching is fine
)
So the guy with the relay box at 0:04 is still PERSON_01 when he comes back at 0:29.
Step 3: kill the false alarms. YOLO flagged a hedge as a person and a reflection on the windscreen as another one. Claude Code caught both and filtered them out:
if track.static_frames > 200 or track.inside(car_glass):
drop(track) # bushes don't steal cars
Step 4: the rule that turns a person into a thief.
if car_unlocked and not owner_nearby and person_at_door:
track.label = "THIEF"
save_clip(t - 10, t + 30)
alert("VEHICLE THEFT IN PROGRESS")
4 steps. 2 thieves. 15 logged events. 0 humans watching the monitor.
Quick note: the boxes and tracking are real YOLOX + tracker output on this footage.
The camera saw everything. The owner slept through it.
That’s the gap Claude Code can close.
Would you let Claude Code watch your car at night? ↓
£25,000 BMW WAS STOLEN IN UNDER A MINUTE. THE SECURITY CAMERA CAUGHT EVERYTHING.
Holy sh*t. Two guys, one keyless relay box, 40 seconds. No smashed glass. No keys. The owner found out the next morning.
So I gave the footage to Claude Code with one job:
“Watch the driveway. Track everyone. Tell me the second it turns into a theft.”
No GPU. No cloud. No $5k security system. One laptop CPU.
Watch the screen:
[0:02] Two blue boxes. PERSON_01, PERSON_02.
[0:04] PERSON_01 pulls out a handheld device. Status flips to SUSPICIOUS.
[0:11] The car unlocks. No owner anywhere near it.
→ Both boxes turn red. PERSON → THIEF.
[0:35] Engine starts. VEHICLE THEFT DETECTED. Footage saved.
Here’s what Claude Code actually wrote.
Step 1: find every person in every frame.
frame = letterbox(caр.reаd(), 640)
out = yoloх.run(None, {"images": frame})[0]
people = out[out[:, 4] * out[:, 5] > 0.12] # class 0 = person
boxes = nms(people, iou=0.45)
1,228 frames. 2,214 detections. Night footage, motion blur, hoodies.
Step 2: don’t lose them. This is where most trackers break. A guy walks out of frame, comes back, and gets a new ID like he’s a stranger. Claude Code added this:
# he left the frame. don't give him a new ID
same_guy = (
gap_s < 4
and dist(last_seen, new_pos) < 260
and new_h > 0.45 * last_h # crouching is fine
)
So the guy with the relay box at 0:04 is still PERSON_01 when he comes back at 0:29.
Step 3: kill the false alarms. YOLO flagged a hedge as a person and a reflection on the windscreen as another one. Claude Code caught both and filtered them out:
if track.static_frames > 200 or track.inside(car_glass):
drop(track) # bushes don't steal cars
Step 4: the rule that turns a person into a thief.
if car_unlocked and not owner_nearby and person_at_door:
track.label = "THIEF"
save_clip(t - 10, t + 30)
alert("VEHICLE THEFT IN PROGRESS")
4 steps. 2 thieves. 15 logged events. 0 humans watching the monitor.
Quick note: the boxes and tracking are real YOLOX + tracker output on this footage.
The camera saw everything. The owner slept through it.
That’s the gap Claude Code can close.
Would you let Claude Code watch your car at night? ↓
THE STORE ALMOST LOST EVERYTHING. CLAUDE CODE SPOTTED THE THREAT. POLICE WERE CALLED JUST IN TIME.
Holy sh*t, security cameras are starting to actually understand what they’re looking at.
Watch the guy in white first. Now watch the back of the store.
Seller’s chilling on his phone. Another guy walks in, goes straight to the back and disappears between the boxes.
Nobody’s watching the monitors. Classic.
So I gave Claude Code the footage and one job:
“Watch the store. Track everyone. Ping me when something’s off.”
It wrote one script, analyze_video.py, and ran the whole thing on a CPU. No GPU, no cloud cameras, no $5k security system.
The setup:
→ OpenCV goes through all 376 frames one by one
→ YOLOX spots every person on every frame
→ ByteTrack gives each person an ID and holds it
→ a storage zone marks where customers shouldn’t hang out
→ activity_events.json logs every move with a timestamp
Here’s the part that broke my brain.
At 0:04 the guy vanishes behind the boxes. Most trackers just give up and call him a new person when he shows up again. This is the line that stops that:
# he ducked behind boxes. don't give him a new ID
same_guy = (
gap_s < 12
and dist(last_seen, new_pos) < 260
and new_height > 0.5 * last_height # crouching is fine
)
So at 0:08 he pops back up and it’s still ID_01. It never lost him.
And the alert rule is literally 2 lines:
if dwell_in_storage >= 5 and seller_still_seated:
alert("REVIEW STORAGE FOOTAGE")
Here’s what the panel actually logged:
[00:03] PERSON_01 → STORAGE AREA
[00:04] TRACK LOST → BEHIND BOXES
[00:08] TRACK RECOVERED → SAME ID_01
[00:08] 🚨 ALERT → REVIEW STORAGE FOOTAGE
[00:10] PERSON_01 → BENDS OVER CARTONS
[00:14] PERSON_01 → HOLDING SMALL ITEM?
Pause at 0:14. Look at his hands.
The seller? Never looked up once.
Oh and the wildest part. It flagged 2 more people I didn’t even notice. One behind the glass on the right, one at the far left edge. Watched this clip 5 times myself and missed both. Look for the thin blue boxes.
4 people. 18 events. 1 alert. 15 seconds of footage.
Now here's where sh*t gets interesting. What happens after the alert?The footage gets saved automatically, the AI gets pinged and instantly analyzes what's going down. And if something seriously sketchy is happening? It calls the cops.
Next version gets one more check:
and track["authorized"] is False
The AI is already trained to spot suspicious behavior and figure out when something sketchy is going down. It could even be an employee secretly stealing products. The system catches the moment, records the evidence, and flags it for review. No random accusations. Just actual footage and decisions based on what's really happening.
No more staring at 10 camera feeds all day.
This video shows the tracking system in action and how AI could help stop a robbery before it's too late.
Cameras used to just record what happened.
Now they catch it while it’s happening.
Would you let Claude Code watch your store? ↓