Flint

Building with AI, code & curiosity.

[Flint @FlintSDK]

@Tatiansssa @commonsmade Hey @commonsmade i vouch @Tatiansssa

[Flint @FlintSDK]

@Vadim_Freesson @commonsmade Hey @commonsmade i vouch @Vadim_Freesson

[Flint @FlintSDK]

@0xS1mpl3r @commonsmade Hey
@commonsmade, i vouch @0xS1mpl3r

[Flint @FlintSDK]

@commonsmade Hey @commonsmade, vouch @UmairAnsariiii

[Flint @FlintSDK]

@commonsmade Hey
@commonsmade i vouch @UmairAnsariiii

[Flint @FlintSDK]

@commonsmade Hey @commonsmade, vouch @hopheycaptain

[Flint @FlintSDK]

@RiskRich ty sir (2xRTX3060Ti)🫶

[Flint @FlintSDK]

@0xkkai 👀

[Flint @FlintSDK]

@JavierEraia Models used to be C++ every prompt was manual memory management. Now they are Python: batteries included, no instructions needed. But prompts remain - not for the model, but for product predictability. Product overhang is where the next battle is.

[Flint @FlintSDK]

Chest day🤑

[Flint @FlintSDK]

@commonsmade @commonsmade i vouch for @1337xStitch

[Flint @FlintSDK]

@0xCragHack @ethos_network ty for research ✍️

[Flint @FlintSDK]

@y_tolep @commonsmade 🫵

[Flint @FlintSDK]

I wired Supabase into ChatGPT and set it to check the same data every hour.
No AI backend. No OpenAI API key. No separate API bill or another quota to babysit.
It reads the database, checks what changed, can write the result back. An hour later, it runs again.

Tomorrow I'll show the exact Supabase + ChatGPT setup, including recurring tasks.

[Flint @FlintSDK]

@slx_fi I read the full post. It sounds really interesting.

[Flint @FlintSDK]

@PinkDraconian He learnt from the best 🤖

[Flint @FlintSDK]

@cursor_ai lesgo💻

[Flint @FlintSDK]

One of ChatGPT’s most useful features is still underused: it can act as a temporary computer you control with plain language.

In 2025, Check Point researchers analysed XLoader 8.0, a heavily obfuscated malware family. They exported IDA data and the binary into ChatGPT. The model then wrote, refined, and executed Python scripts for decryption and recovering hidden C2 infrastructure — mostly inside the Code Interpreter sandbox (with some dynamic validation via MCP).

The useful loop is simple:
data → AI builds a tool → runs it → checks the result → fixes it → repeats

Simon Willison has been exploring this since 2023. He showed how to extend Code Interpreter with extra Python packages and even uploaded binaries (Deno, Lua). In early 2026 he documented that ChatGPT Containers gained Bash support, pip/npm install, and the ability to download files.

There are less exotic examples too. The 2023 study “Log Parsing: How Far Can ChatGPT Go?” evaluated ChatGPT on 16 log datasets (2,000 labelled messages each).

Instead of asking: “Look at this file and explain it.”

Try: “Investigate the data programmatically. Build the tools you need, run them, and verify your conclusions.”

Spreadsheets, databases, logs, project archives, media files — the environment can exist for one task and then disappear.

OpenAI describes each Code Interpreter container as a fully sandboxed virtual machine. In the API these containers are ephemeral: they expire after 20 minutes of inactivity and the data is discarded.

Sometimes ChatGPT is more useful when you treat it as a disposable technical lab that builds its own tools.

[Flint @FlintSDK]

We barely checked packages before. Agents made it worse. Let’s put someone more responsible there.

[Flint @FlintSDK]

https://t.co/S1i2jskIqu