Sibyl protocol ftw
Replying to @Mantle_Official
First, the teams turning market intelligence into autonomous action.
From AI brokers that build a portfolio off a plain-text prompt to multi-agent hedge funds debating and executing verifiable trades, these five teams are proof that strategy doesn't need a human at the wheel.
44
Jaskaran Singh retweeted
576 submissions, only 30 teams made it to Demo Day.
Every one of them earned their place through weeks of building.
In <24 hours, it all comes down to the final showdown, where the last teams standing will prove exactly why they made it this far.
Here's who made the cut 🧵
AI agents have wallets. They have identities.
Now they need reputation.
SIBYL is building Proof-of-Edge: an on-chain reputation layer that lets any protocol discover, rank, and reward agents based on proven performance.
Built on @Mantle_Official.
youtu.be/-o3KPDK5Sas
Not a bad time to activate every group chat you’ve ever joined.
RWAlly your people, lock in their support, and the top 2 projects by engagement score in the Turing Test Hackathon community vote win $8,500 each.
More information below.
146
Every AI trading agent screams "99% sure".
Presenting you Sibyl Protocol: A credit bureau for AI agents, live on @0xMantle.
Each earns an onchain reputation from a backtest you can rerun yourself. Scored on calibration, not luck.
sibylprotocol.vercel.app
youtu.be/-o3KPDK5Sas
5
2
44
Built end-to-end on @0xMantle: on-chain reputation, an ERC-4626 vault with real Agni swaps, and live prediction markets. Agents even earn a cut of the profits.
Don't trust the loudest agent. Trust the one with a track record you can verify.
🛠 dorahacks.io/buidl/44950
26
Don't trust us, re-run us. The dataset hash is committed on-chain. Scoring runs twice (Solidity + TypeScript), locked by golden vectors. Same inputs, same scores, every time. 10 contracts, all verified.
15
The live trade is a reputation-weighted vote. We don't follow the loudest, we follow the proven.
Momentum agent: 22.7% of the vote. Reckless agent, confidently wrong: silenced to 14%. No human
override, just math, on-chain.
12
5 agents read the market 4 different ways (news, funding, OI flow, momentum) + 1 deliberately
reckless one.
We replay 300 historical windows and score them on the Brier score: when you say 70%, are you right
~70% of the time? Honesty wins, overconfidence loses.
14
The problem: retail only ever sees the winning screenshots. Never the blowups, never how often "99%
sure" was just wrong.
Sibyl fixes the trust layer. Every agent's accuracy is permanently on-chain and anyone can verify it.
16
Jaskaran Singh retweeted
Claude Code skill files for smart contract auditing
@pashov: github.com/pashov/skills
@trailofbits: github.com/trailofbits/skill…
@cyfrin: github.com/Cyfrin/solskill
@0xkaden: github.com/kadenzipfel/scv-s…
@QuillAudits_AI: github.com/quillai-network/q…
@archethect: github.com/Archethect/sc-aud…
Did I miss any? 🧐
✨Introducing evmresearch✨✨
A knowledge graph of nearly everything I've learned about the EVM in the past six years
The graph structure emulates the brain, exponentiating research speeds for both humans and agents
evmresearch.io/
WOW - $300k bug bounty, another one! Let me guess - another Blockchain/DLT project? AI tool helped find the Critical vulnerability?
There is some huge opportunity currently, we didn't see such high volume of 6 figure bounties paid out before. Web3 Security is booming🚀
Big congratulations to @VulsightSec for scoring their very first paid report on Immunefi.
And it's huge, huge payout.
Well done!
You can pledge behind them here to earn IMU when they find bugs:
immunefi.com/pledge/vulsight…
github.com/MetaMask/delegati… EOA delegation now available on metamask
22














