🕯️codeislight.eth 🕯️

Founding Engineer at ❓ | also @SwapXfi ⚡️|
old account @codeislightEth ⏮️

[🕯️codeislight.eth 🕯️ @codeislight]

🧵+12M in "stuck" crypto might be recoverable.

As long as the deployer nonce on another chain is not burnt and the private key is still around. The recovery mechanism is based on CREATE opcode in the EVM.

Special thanks to @pcaversaccio for help validating this finding.

1/9

[🕯️codeislight.eth 🕯️ @codeislight]

@DanGallagherDC

[🕯️codeislight.eth 🕯️ @codeislight]

@solidintel_x Just use @robinscanio

[🕯️codeislight.eth 🕯️ @codeislight]

@0xConda Federal reserve must be sleeping on this

[m4rio @m4rio_eth]

everyone check your rust codebases!!! short prompt

Check <repo> and this machine for the Aug 2026 crates .io compromise:
arrayref 0.3.10, append-only-vec 0.1.9, proc-macro1, proc-macro-en.
Grep every Cargo.lock/toml, then check ~/.cargo/registry for those crates, /tmp/rust-setup, egress to 23.254.165.112, and new LaunchAgents
since 2026-08-19. Report only, don't fix.

[🕯️codeislight.eth 🕯️ @codeislight]

@bailsecurity Wow, that's a day and night difference!

[🕯️codeislight.eth 🕯️ @codeislight]

@RikispRikisp They're goated!

[🕯️codeislight.eth 🕯️ @codeislight]

@0xCharlesWang that's pretty dope!

[🕯️codeislight.eth 🕯️ @codeislight]

SC-AI assessment by @bailsecurity has been incredible. ahead of our main audit it spotted serious issues, and the act of resolving them made us reflect on and revise key design decisions and simplify modules that could have otherwise been problematic. By far it's the best AI auditing tool out there I have experienced so far thanks to the extent and depth it can analyze that a manual review might miss. We ran it over multiple rounds, roughly a day per module and it kept surfacing findings worth fixing. Running it recursively is what really paid off, audit a module, fix it, run it again, with fewer findings each pass until they came back clean. I believe it's a must-have as a preflight assessment before any serious audit engagement.

[BailSec @bailsecurity]

Some of our partners ran their codebases through every major AI auditing tool available.

Then they tried ours.

The delta was... uncomfortable.

All our partners and us will share the results very soon.

[🕯️codeislight.eth 🕯️ @codeislight]

@apoorveth

[0xBunny | ... @zeroxbunny]

https://t.co/eHM71UrZfr

[🕯️codeislight.eth 🕯️ @codeislight]

@ProgrammerSmart Ohhh that's really clever! So it's exponentially iterative

[🕯️codeislight.eth 🕯️ @codeislight]

@0xngmi since a decent amount (30m$+) was tipped off to @titanbuilderxyz validator, is there any possibility of compensation?

[🕯️codeislight.eth 🕯️ @codeislight]

@apoorveth @Rabby_io 🫩

[🕯️codeislight.eth 🕯️ @codeislight]

just noticed that Multicall3 first and last 2 bytes are mined to have "Call" lol

https://t.co/oCG8U1EO04

0xcA11bde05977b3631167028862bE2a173976CA11

[🕯️codeislight.eth 🕯️ @codeislight]

@cz_binance @BNBCHAIN Majority are bots tho 😂

[🕯️codeislight.eth 🕯️ @codeislight]

@SECGov 👏

[🕯️codeislight.eth 🕯️ @codeislight]

@ProgrammerSmart Yaaay!

[🕯️codeislight.eth 🕯️ @codeislight]

@apoorveth @inversebrah Greatest thing i've seen on CT so far

[🕯️codeislight.eth 🕯️ @codeislight]

@apoorveth @notnotstorm this is the kind of double counting that matters :)

[Avi Chawla @_avichawla]

AWS did it again!

They have introduced a novel way for developers to build Agents.

Today, when you build an Agent, you start with a simple goal, then end up juggling prompts, routing logic, error handling, tool orchestration, and fallback flows.

One unexpected user input and the whole thing collapses.

Strands Agents framework by AWS approaches Agent building differently.

It takes a model-driven approach that lets the LLM decide how to plan, choose tools, and adapt to edge cases on its own.

You provide the capabilities and guardrails, and the model handles the workflow, which is different from the brittle agent frameworks we typically use.

Here are the highlights.

- TypeScript SDK
> You can build Strands agents in both Python and TypeScript. The TS SDK gives a modern async-friendly workflow and works with any provider. Amazon Bedrock, OpenAI, Anthropic, you name it. Swap models or providers without touching your agent code.

- Edge device SDK
> You can run agents directly on small devices using local models, with bi-directional streaming.

- Strands steering
> Instead of writing massive prompt blocks to force an agent into a workflow, Steering lets you guide it toward a desired outcome. The model handles the steps dynamically, which keeps agents flexible and reduces token use.

- Strands evals
> A practical evaluation suite that lets you test agent behavior, compare versions, catch regressions, and validate correctness or safety before deploying. This is critical as agents move from demos to production.

Everything is fully open-source!

One particularly interesting thing is that Strands pairs with AgentCore, AWS’s secure platform for running agents at scale.

So you can prototype with Strands in a few lines of code, then deploy in AgentCore with built-in policies, monitoring, and real-world evaluations.

This is important because:

- You build workflows that adapt rather than break.
- You choose the models and providers you want, without rewrites.
- You can run agents on cloud, on device, or inside your own network.
- You can evaluate, refine, and trust them before putting them in front of users.

You can find the Strands Agents GitHub repo in the replies!

If you want a hands-on demo showing how to build a Strands agent and deploy it to AgentCore, tell me, and I’ll record one.

Thanks to AWS for building this and working with me today!