Derek Heinrichs

Founder @ HSS | Smart Contracts, Blockchain | Building infrastructure for autonomous AI agents | x402 | Anthropic CVP-approved

[Derek Heinrichs @DH_OnChain_dev]

Someone who's opinion I highly value asked me about agentic fleets. I felt compelled to answer their question, but then a light clicked in my head, what if I could extract the frame of my fleet for others to use?? Funny enough I actually applied to the XRPL Commons Paris event in September for this exact thing but I hadn't built a replicable system for someone to use until now. So, even though I'm technically shooting myself in the foot here, I present to you, open-sourced MIT licensed and free to use, my own personal agentic fleet framework built from my wonky mind,

https://t.co/uHT3Jk2Kx8

You can do whatever you want with it, make sure to read the Readme .md it thoroughly explains what I've been perfecting for about 3 months now. Enjoy!!! Its viable for any project, and I highly recommend following the phases, if you can improve it please do I'm all ears!! Now everyone can have an agentic fleet of their own.

If you have any trouble using it please let me know.

Incase I can't get to you fast enough, you can clone the repo, and open Claude code in a parent folder and have it do a /init to build what the project does and then ask it to integrate your work.

[Derek Heinrichs @DH_OnChain_dev]

built an agentic fleet that pinpoints solar-ready land across Florida. (Claude code is addicting in /rc)

It ranks parcels on what actually makes or breaks a project, proximity to interconnection and the capacity left on that line, contiguous buildable acreage after wetlands and floodplain come out, and land use that clears without a multi-year zoning fight. The dials are all there, search radius, minimum megawatts, distance to substation, and the model sorts the whole map down to the parcels worth real diligence. If you touch Florida solar land development in any capacity, comment below or follow me and I’ll follow back. I’d rather compare notes with people building this for real than post into the void.

[Derek Heinrichs @DH_OnChain_dev]

@JoelKatz @lukestamps Raisins have their place, just not in cookies or carrot cake.

[Derek Heinrichs @DH_OnChain_dev]

@_JackMcDonald_ @Mastercard Always believed. $RLUSD is a really cool addition!!

[Mastercard @Mastercard]

Today, Mastercard is announcing plans to expand settlement capabilities to include stablecoin, intraday, holiday, and weekend options, giving partners more choice in how and when transactions are settled. That means we’re:

✅ Enabling greater choice to settle in fiat or regulated stablecoins

✅ Improving liquidity management for time sensitive, cross border flows

We’re supporting settlement with @Circle’s USDC, @Paxos-issued stablecoins including USDG,USDP and @PayPal’s PYUSD, @Ripple’s RLUSD and @SoFi’s SoFiUSD across a range of supported blockchain networks including Arbitrum, @Coinbase’s Base, @CantonNetwork’s Canton, Ethereum, @0xPolygon, @Solana, @Tempo and XRPL. ARQ Finance, CBW Bank, @crossriverbank, @Lead_Bank and @Nuvei will be among the first to support.

[Derek Heinrichs @DH_OnChain_dev]

you are so right! i just wonder what that shift will be?

this is what im seeing, the model is never waiting on compute, just me!

more access doesnt make me wait less, it just branches the work into parallel streams with me as the bottleneck.

i read a chip like this less as my tasks finishing faster and more as me being able to run far more of them side by side, horizontal scaling for people sort of.

do you think there is a max capability velocity where a single person cant feed or review any faster and past that point the only move is to hybridize and take on more tasks instead?

seeing all of us rapidly, exponentially increase our productivity and accuracy/precision has been wonderful! the optimist in me want to believe it gets even better.

[CoinDesk @CoinDesk]

NEW: @Mastercard expands settlement to support regulated stablecoins including $USDC, $RLUSD and $PYUSD across Ethereum, Solana, Base, Arbitrum and XRPL.

[Derek Heinrichs @DH_OnChain_dev]

“Unfortunately, after careful consideration, we will not be progressing with your candidacy for this role.”

This is why I stick to BB/audits on blockchain. Read mastering Ethereum, scan contracts/repos cherry pick promising bugs and submit the reals. Never have to mess with hr again at this point.

[Derek Heinrichs @DH_OnChain_dev]

@JoelKatz @KPMorningstar “Woe to you, oh earth and sea
For the Devil sends the beast with wrath
Because he knows the time is short
Let him who hath understanding
Reckon the number of the beast
For it is a human number
Its number is six hundred and sixty six”

[David 'JoelKatz' Schwartz @JoelKatz]

14 years ago, we got together with an idea to build a better way to move value. What happened next was something none of us could have built alone.

And by "us," I don't just mean the three of us.

I mean the developers, validators, businesses, community members, and everyone who helped shape XRP into what it is today.

Happy Birthday, XRP!

[Derek Heinrichs @DH_OnChain_dev]

@gpt3_eth @jamesdouma Friction at zero you say?? I’m so down for it! Friction removed from thoughts and money sounds like a global golden age to me.

[Derek Heinrichs @DH_OnChain_dev]

@WatcherGuru I wonder if the volatility in crypto will
be quelled by agentic traders. Hopefully with stop losses baked in.

[Derek Heinrichs @DH_OnChain_dev]

@dom_kwok @kwok_phil Lfg!

[Derek Heinrichs @DH_OnChain_dev]

🛡️ Spent the last week in the HSS Security Lab building something I'm genuinely proud of: an autonomous vulnerability-research engine that hunts smart-contract bugs 24/7 across Immunefi, Cantina, Code4rena, Sherlock, Hats and CodeHawks.

Here's the pipeline:

→ Continuous discovery + scope-resolution across 6 bug-bounty platforms

→ Multi-agent analysis on Claude Opus — source-grounded, archetype-aware

→ Adversarial verification: every finding faces independent agents whose only job is to disprove it
→ A deterministic, regression-tested gate suite that runs before anything reaches a human

And the lesson that ended up shaping the whole build:
The hard part of automated security research isn't finding suspicious code. It's reliably discarding the ~95% that's wrong.

An LLM will hand you a dozen "criticals" an hour. Most are mirages like an oracle check that's actually present, "theft" that quietly returns funds to the protocol, a bug the project already documented in its own repo.

So most of the week went into reliability, not recall:

• A proxy-resolver that follows upgradeable proxies to their real implementation, scan the delegate shell and you're scanning an empty contract

• A known-issue gate that reads a project's own audit & fuzzing docs before it ever flags a finding

• A 24/7 deployment-watcher that catches freshly-deployed in-scope contracts within hours because fresh code is where the bugs still live

Automated security is a grind of false positives and it is incredibly important to filter out the hallucinations.

Building in public, more soon. 🔍

#SmartContractSecurity #Web3Security #AI #DeFi #BugBounty #SecurityEngineering #BuildInPublic

[ClaudeDevs @ClaudeDevs]

We've reset 5-hour and weekly rate limits for all users on Pro and Max plans.

We fixed an issue that caused some Claude Code sessions to spawn excessive parallel subagents, burning through usage faster than expected.

[Derek Heinrichs @DH_OnChain_dev]

@gpt3_eth @forgebitz Neck just about snapped

[Derek Heinrichs @DH_OnChain_dev]

Reminds me of when everyone said crypto and blockchain had no utility before ai agents were able to use crypto wallets with x402. Watching visa/Mastercard/stripe join was just icing on the cake.

I’ll take convictions and hard work fixing mistakes/adjusting code > curling up in the corner of the room.

[Derek Heinrichs @DH_OnChain_dev]

@RanjYousif @Polymarket It’s not entirely true, most chatbots have safety rails, if you cross them by asking to build illegal code you get flagged. So yea they could try but most likely it won’t be useful.

[Derek Heinrichs @DH_OnChain_dev]

Attack vectors built by ai are only good against systems that haven’t been audited by ai. It’s not great at making new unique attack vectors, fortunately that’s the human strength.

Will this remain true for much longer?

That I’m not too sure of, if Anthropic a Mythos is anywhere close to what opus 4.8 ended up being I think we are okay for awhile though 🤣🤣 already miss opus 4.7 about 2 weeks ago.

[Derek Heinrichs @DH_OnChain_dev]

@gpt3_eth @t_blom I feel like all I do now is train ai, but I’ll admit, training gpt 5.5 or opus 4.8 isn’t too bad of a job 😆

[Derek Heinrichs @DH_OnChain_dev]

@jessepollak @base Commerce is evolving with ai and crypto, based. @base