Silwolf

[Meta Alchemist @meta_alchemist]

continuing to train my specialized agents with verifiable loops with Spark Loops desktop

pushed a demo script producer agent from 62% to 89%, scored by GLM 5.2, Deepseek v4, and Grok 4.5 yesterday

here is a deep dive:

[Meta Alchemist @meta_alchemist]

There are many ways to train an agent. Most of them let the agent grade its own homework.

Spark Loops doesn't. It uses a multi-LLM framework built so agents can't cheat their way through training.

The main LLM edits and mutates its domain knowledge chip to solve tasks better, but it never scores itself.

That job belongs to the reviewer LLMs. They score each run and hand back feedback.

The main LLM takes it, restructures how it thinks about the domain: its processes, its problem-solving moves, adopts what works, and goes again.

New run, new feedback, new mutation.

Reviewers: they never talk to each other.

Each one is blind to the others' feedback, so every score comes from independent reasoning and judgment with no room for groupthinking or echo chamber.

The result: an agent that gets better in the domain you want it to improve upon because it earned it, not because it gamed the grading system.

You can choose who you want your main LLM to be, and who you want your reviewer LLMs to be.

I do my runs with 1 main and 3 reviewers, enough perspectives to keep the feedback comprehensive.

Since I'm gonna be doing more Spark Loops product videos, I created a new domain chip for scripts today. Let's see where we can take it.

[Meta Alchemist @meta_alchemist]

Loop engineering and training your own agents with verifiable loops is about to get easy for everyone.

Here is the first look to Spark Loops desktop app, where you can:

> create your agents
> train them without any technical friction
> use them in sessions and your workflows

[Meta Alchemist @meta_alchemist]

They locked the frontier.
Too dangerous they said...

But open source builders broke their rules.

[Meta Alchemist @meta_alchemist]

Spark R29 is now live!

R29 consolidated accepted work across Spark CLI, Telegram Bot, Builder, Spawner, Researcher, Character, Voice, Personality, Memory, Domain Chips, and the installer surface from community PRs.

The biggest changes:

Spark CLI is safer.

R29 added stronger approval gates for risky commands, module-name sanitization, SQL allowlists, Unicode prompt-injection normalization, atomic SSH known_hosts writes, npm install-command restrictions, and better secret/path redaction.

Telegram is more reliable.

Accepted fixes improved group command parsing, API-key/stderr redaction, SQLite WAL shutdown, conversation memory races, temp-file cleanup, socket cleanup, preview URL SSRF checks, and slash-command handling.

Builder is harder to break or abuse.

R29 added provider SSRF validation, OAuth transaction safety, minimal child-process environments to avoid secret leakage, SDK module allowlists, JSON guards, and safer runtime path behavior.

Spawner got mission stability work.

R29 added lifecycle dedupe, retry tolerance, timeout settlement, reconnect jitter, scheduler in-flight dedupe, workspace containment, event cleanup, and constant-time comparison hardening.

Researcher got safer execution paths.

Accepted work covered path traversal prevention, SSRF/IP validation, stale lock recovery, subprocess timeouts, corrupt JSON guards, and public summary redaction.

Character, Voice, Personality, and Memory got durability and privacy fixes.

That includes .env exfil detection, voice SSRF protection, NaN/Inf rejection, atomic personality writes, path traversal guards, corrupt YAML warnings, log growth bounds, and safer profile loading.

The release also included smaller but important system fixes across Spark-Agent-Site, Domain Chip Labs, Domain Chip Memory, and Startup Bench.

R29’s theme: fewer leaks, fewer crashes, safer local authority, and a more reliable agent path from install → Telegram → Builder → mission → memory/persona/voice.

Spark Compete leaderboards are updated as well. Still there are more community PRs to ship in the next installers, so the current leaderboard is not final.

See you in the next updates!

Make sure to update your Spark agents 🫡

[Meta Alchemist @meta_alchemist]

Ayyy pirates of thr Seas! 🦜

Seas of Spark waitlist now has over 6000 pirates. I'm expecting to get the playtest build ready in 3-4 days.

From the first playtest everyone will be able to earn items that persist through seasons through a mechanic called Beacon of Light

Whereby you'll spend the gold you make via beacons, and with each beacon you'll have a RNG chance of dropping items, ships & pets that give your certain boosts.

Also, there will be certain items where you can rush your beacon level to required levels and pay in gold to get limited stuff, in a non-RNG based way.. These items will be tradeable/sellable in marketplaces too and will be sought after due to boosts they give in-game.

If you haven't joined Seas of Spark waitlist you can do it through here: https://t.co/IoYtx7qYa6

Make sure to invite friends into your crew via refs for multipliers, and a higher stance on the waitlist.

Ahoy 🏴‍☠️

[Meta Alchemist @meta_alchemist]

2500 pirates already joined the waitlist for Seas of Spark, thank you all 🫶

[Meta Alchemist @meta_alchemist]

Today I want to talk about the economy I envision in Seas of Spark a bit.

What I want to create here is a full on MMO economy, with certain items having real life value.

Just like you saw in games like Runescape or WoW but more like having the rails to do this properly.

I want people to really feel the excitement of progression, survival, PvP, finding a good loot, getting a great chest, having the right skill tree that creates a perfect combo, within a world of Seas, fishing, automation and AI.

Blending genres, while creating new form of gameplay systems through having an agent orchestrate your fleets, fishing village, and more when you are not online too.

While bringing into the game the risky stuff like full-loot PvP mechanics on regions with better loot and more monstrous fishes and collectibles.

Ai giving the full ability to blend everything I loved from games of RTS, Survival, and MMO games.

Red Alert 2, Age of Empires, Rust, Conan Exiles, Ultima Online, World of Warcraft, Albion Online, and more games that shaped my online gaming past times.

So, expect something more than fishing as we roll out things phase by phase at Seas of Spark...

[Silwolf @silwolfy]

Join the Seas of Spark playtest with me, a fishing MMO with limited NFT collectibles available for leaderboard spots: https://t.co/O0BSnmST2D

[Meta Alchemist @meta_alchemist]

LIVESTREAMING

Vibe coding a fishing survival MMO game - Seas of Spark https://t.co/ZTCgaw4Cq0

[Meta Alchemist @meta_alchemist]

Seas of Spark first playtest is around the corner. People are starting to get their tokens to get ready!

Under QRt'ed post you can drop a 🏴‍☠️ for a chance of Founder Ships.

[Silwolf @silwolfy]

@meta_alchemist 0x48a2e66e5036313b5c0f77c8d09a321abf17a50b 🏴‍☠️

[Meta Alchemist @meta_alchemist]

When vibe coding a game one of the most time consuming part is to lock down a coherent art style.

My goal with Seas of Spark is to have such workflows that I can then input into Spark agent as rulesets so I can get other people who'll vibe code benefit from what takes much longer into agentic loops and workflows.

The good part about Spark is, it works with something called domain chips, that can train on workflows, and get better at them.

If we can get agents to do the slow parts faster, and with higher quality, then everyone can make better games.

So while I build Seas of Spark, I get a lot of stuff that will later benefit everybody that uses Spark in similar objectives.

I've been working on PvP systems today, and already getting something ready for the first version of the game.

[Spark @Spark_coded]

Let's stream Building with Spark sessions, every weekday for 12 hours this week?

[Spark @Spark_coded]

A new ruleset:

If you have gotten community airdrops (not talking about Seedify/Seedworld/Vibecoder)

And if you don't hold at least half of your tokens, no future tokens will be issued.

The rest will be given to new community members through new events, who'd rather hold a portion and become part of the community.

This is so we can create a real community and ecosystem rather than just rewarding tourists.

[Silwolf @silwolfy]

@meta_alchemist @Spark_coded I'm in 0x48a2e66e5036313b5c0f77c8d09a321abf17a50b

[Meta Alchemist @meta_alchemist]

It was more than 5 years ago that I launched my first token as an airdrop. It was a stressful experience, not knowing where it would go, whether I could handle it.

I didn't have more than $5k for marketing, all the money was gone to liquidity and airdrop fees, and 30k+ people were in the airdrop.

The launch day was pretty exhausting during creating the liquidity and sending the tokens.

Later I was shocked that it started higher than the initial price we've set it, and then few days later it went higher, and then much higher.

Then we saw the first drop, I felt like a train was moving, and I couldn't get off of it.

I had to fight back.

Then miracles happened, what started from nothing just started to act like it had some rockets.

9 months later we were sitting at $350 million marketcap, outcompeting all the launchpads, and there were like 100+ but we had it in us.

In September, a storm came to take it all, and it partially did, but it couldn't take something I learned along the way: never quitting, never surrendering.

Today, I'm starting Spark. Same vibes. Not knowing where it will lead, but something in me says, much higher.

Again, an airdrop launch, not much money after the hack took it all from me. Last money again, being put into liquidity and airdrops fees.

This time, no marketing spend, just AI subs. I got 8 Codex 20X, 1 Kling, and 1 Higgsfield.

Let's run it back, turbo!

[Spark @Spark_coded]

// LAUNCH TIME
4 PM UTC

// MISSION
SPARK THE BASE

[Spark @Spark_coded]

Grants, hackathons, and incubation will grow the Spark ecosystem while providing free tokens to all holders.

a self-improving agent
meeting
a self-improving ecosystem
meeting
a self-improving portfolio

Get Spark'ed

[Spark @Spark_coded]

Just a day left from TGE at @base

Who is excited?

🙋‍♂️