THIS GUY GOT TIRED OF MANAGING AI AGENTS THROUGH TERMINALS AND DASHBOARDS SO HE BUILT THEM AN RPG WORLD
5 agents and each one has a pixel character, a station, and they actually walk around the space
when enough unresolved issues pile up, the agents walk to a meeting point and hold a council session.
four different models debating what to do next, not scripted. each one reads the live system state independently.
in one session an agent pushed for cold outreach to close leads at 2am. another one said that's a terrible look for an autonomous system contacting strangers while the operator sleeps.
they ended up pivoting to an inbound strategy that none of them originally proposed.
single HTML file, node bridge, and phaser. runs on a Mac Mini.
instead of reading logs and checking dashboards you just watch your little pixel agents walk around and talk to each other
this is the most creative way i've seen anyone manage AI agents so far
After setting up Claude code + Tradingview
first thing I built is a bottom detection indicator.
I've been manually eyeballing RSI divergences + volume spikes + key support levels to try and catch local bottoms.
works sometimes. misses a lot.
So I told Claude:
Write a Pine Script indicator that detects potential local bottoms using:
> Bullish RSI divergence on the 4H (price making lower low, RSI making higher low) > Volume spike above 2x the 20-period average > Price within 1% of the lower Bollinger Band (2.5 std dev) > Stochastic RSI below 20 and crossing up
When all 4 conditions align, plot a green diamond below the candle and fire an alert.
Claude wrote it. Injected it into TradingView.
First compile had errors with the divergence detection logic. Claude read the errors, rewrote the lookback function, recompiled. Clean.
Then: "Run it on BTC 4H and screenshot the last 3 months. How many signals did it fire? How many were actual local bottoms?"
Claude read the chart back:
→ 7 signals in 3 months → 5 hit within 2-3% of the actual local bottom → 2 were false signals that kept bleeding
Not bad for a first version.
"Now tighten it. Add a condition that volume must be increasing for at least 3 consecutive candles before the spike. And add an EMA 200 filter only signal when price is above the 200 EMA on the daily."
Recompiled. Reran. False signals dropped to 0.
Total signals dropped to 4 but all 4 were within 1.5% of the local bottom.
Then I ran it across $SOL, $ETH, and $HYPE. Same timeframe. Claude compared the results across all 3 and told me which pair the indicator works best on.
The entire process.. idea to working indicator to multi-pair validation took about 15 minutes.
Normally this would be a full weekend project. Writing the Pine Script. Debugging. Manually scrolling through charts.
Tweaking parameters one by one. Checking each pair separately.
I'm not saying this indicator is the holy grail. It's a starting point.
But the speed of going from idea → code → validation → iteration is completely different now.
Claude code doesn't give you an edge. It lets you test your ideas fast enough to actually find one…
“WE HAD A GOOD THING YOU ORANGE SON OF A BITCH, we had congress, we had the house. All you needed was to shut up and deport all the people you wanted but no, you just had to be Israel’s slave”
New in Claude Code: auto mode.
Instead of approving every file write and bash command, or skipping permissions entirely, auto mode lets Claude make permission decisions on your behalf.
Safeguards check each action before it runs.