efe

Solana has had $2B+ drained from programs in the last 3 years. 99% of it happened AFTER the attacker already had a valid keypair. Signing a transaction ≠ authorization. I built something to fix this. It's called Trana. 🧵

It’s been long. What we are building?

Should fairness be a spec?

Grasp what we do in under a minute!

This fires… Confire will going to be of the most seamless and effective AI plugin. It comes in when it needed, most job done silently. Follow and support @confiredev

@xer0c @BilyanaChristov @tributaryso @colosseum Awesome news !!!

Another big win for @tributaryso 🍾🍾 Its a top 10% project! The interview already happend. This was intense (1am) but a great experience meeting the people behind @colosseum.

Follow @confiredev

@SuperteamPOL @mihal_wojtas @Nene_X21 @TradingBattles @solana Welcome Adam

Meet our newest STPL member, @Nene_X21 👋 Straight from Sosnowiec, Adam brings creative business strategy expertise and vibrant live presentations to the community. With 8+ years of experience across Web3 business development and operations, he’s now building Trading Battles (@TradingBattles) - an infrastructure layer for 24/7 competitive trading. Outside of business, he’s into Formula 1 and football, an Avicii enthusiast, and a big chicken & waffles fan. Welcome, Adam!

dev.confire.dev #AI #DevTools #Startups

Built Confire, a context & tool firewall for AI coding agents. Reviews risky calls (tools, MCP mutations) before they run + sanitizes noisy outputs (Bash, Figma, GitHub, etc) before they hit the model. 40-95% noise reduction. Raising angel round. 👇

Most visitors who'd benefit from chatting never send the first message. They see a blank input and close the tab. So we fixed that. Text's widget now reads the page and suggests questions.

@its_axios @pawel_rainer @Helius He is just transparent as always

@pawel_rainer @Helius you really have a knack in giving constructive feedback

I was on @Helius business plan for almost a year and cancelled subscription today. They are extremally expensive (especially for live streaming!), rate limits are misleading (100rps but sometims hitting 429 with 10rps still not sure why tbh) and are "credit based". Switched to @solvibestation (no offence guys but name is terrible😂), but: - 100rps for 75$/mo, unlimited credits - 5 grpc connections for 150$/mo, unlimited bandwidth In general just works

@SuperteamPOL @0xbeys22 @solana @superteam Winners always build: just a reminder

JUST A REMINDER: BUILDERS ALWAYS WIN.

Dayjob, then afternoon job, then night job.

@holapabs I’m treating tool output as something that needs normalization before it reaches the agent. The main filter is: “does this token improve the next action?” So it keys on action relevance, duplication, source authority, transient vs durable info, and whether raw output is optimal.

@holapabs - does this help the agent decide the next step? - does it change implementation/review behavior? - is it source-of-truth or repeated wrapper text? - is it stable context or one-off tool noise? - can it be replaced with a normalized instruction or pointer?

I am running into very good results with agent context optimisation in Claude Code with a methodology I found yesterday. Reduced overall context 30%, and reduced context noise up to 70%, and there is one more result ~95% but this reveals later 👋

.go is good. I like Go

@solana Bball? NBA sponsor?