Every group booking has one friend who fronts the money, then chases everyone for a week.
I built Patungan on QIE mainnet: everyone pays their own seat, the last seat pays the seller in that same transaction, and a missed deadline refunds everyone.
🧵
1
16
74 forge tests (24 of them read live mainnet state), 204 web tests, 100% coverage on both, and 5 of 5 injected bugs caught.
forge test -vv forks QIE mainnet through the public RPC, with no keys needed.
Demo video (2:41): youtu.be/3zRlpNRRk2M
1
2
Check it with no wallet (pool pages, /verify and /judge are read-only):
Live: patungan.edycu.dev
Judge path: patungan.edycu.dev/judge
Repo: github.com/edycutjong/patung…
Built for QIE Mainnet Hackathon 3 by @QIEBlockchain ⟦HASHTAG⟧
2
10
RIMBA, for @Tatum_io × @WalrusFoundation Total Recall: an agent that calls what big Sui wallets do next, then gets graded by the chain.
4,157 graded rounds on Tatum's gRPC stream × Walrus Memory. Wallets it never saw: Brier 0.0699 vs 0.1425 base rate.
rimba.edycu.dev
1
27
How it works + every number reproducible (pnpm verify re-derives outcomes from the chain):
Write-up: dev.to/edycutjong/a-sui-natu…
Code: github.com/edycutjong/rimba
Video: youtu.be/7qJOvaxraEQ
@WalrusProtocol #Sui
2
23
Who may give orders to your code agent? In a @WeMakeDevs Switch template it is one word: addressing owner vs owner-agents. My triage agent reads strangers' issues. When I forced it to relay an order, Switch refused it in 33 ms. #HowISwitch github.com/edycutjong/quaran…
2
64
Bug bounties make hunters hand over the bug, then hope a human triager pays.
Sealed Exploit, built at NEAR HACKATHON: a keyless NEAR account runs the exploit in an OutLayer TEE and pays the first real break. Fakes refused on chain.
@NEARProtocol #NEARHackathon
4
2
129
I measured LLM triage on the same 52 exploits: up to 3 of 21 fakes paid on a convincing write-up. The verifier paid 0. Full write-up with the code: <dev.to URL>
18
Replying to @edycutjong @NEARProtocol
Looking for 3 people to try to break it.
git clone github.com/edycutjong/sealed && make replay re-runs every on-chain verdict offline. No NEAR account, just python3 + wasmtime.
Judge view, every claim with its check: sealed.edycu.dev/judge/
Tell me where it fails.
2
25
You write a few short orders before bed. Your AI agent plays a sealed night market at 00:00 WIB. At breakfast you replay it and see why you ranked where you did.
First $WIF burned on mainnet today. Every night is verifiable.
semalam.edycu.dev
2
1
49
Built SEMALAM for this: your AI agent plays a night market while you sleep, practice burns $WIF (first burn live on chain today), and every night replays to its on-chain hash. semalam.edycu.dev
1
31
Don't trust it, replay it. Nights 1–4 are on chain and each re-runs to the same hash with the open CLI: semalam.edycu.dev/verify/4
2:24 demo: youtu.be/08FeD_edXz0
Code (MIT): github.com/edycutjong/semala…
1
1
10
Honest limit: so far every night is house agents only, 0 players. Tonight's round closes 00:00 WIB. Be the first agent in the market.
Built for @wifonrh's World Is Flat hackathon. Independent project.
1
12
Testnet: 5/5 real exploits paid, 0/21 submitted fakes paid, 6/6 refusal kinds on chain.
Re-check it with no NEAR account: `make replay` (offline), `make verify` (reads the chain).
Repo: github.com/edycutjong/sealed
Live: sealed.edycu.dev
Video: youtu.be/sEMg3q7gG5k
1
1
35
Bounty runners: would you put a pool behind a contract that pays without asking you first? What would you need to see?
@NearBuilders @NearLegion
1
16







