Actually I tried it but it still have many caveats and feels like a experiment and github stars are insignificant.
Why??
1. In short this uses another llm to evaluate the model tasks and updates the skill files.
But claiming as ml weights is still misleading.
🚀 Introducing SkillOpt — an optimizer for agent skills.
Instead of finetuning model weights, we treat a natural-language skill as a trainable external parameter.
Think of it as deep learning for the frontier-model + agent era: learning rate, LR schedule, mini-batch, batch size, epoch, momentum — all in text-space optimization.
SkillOpt enables stable, controllable skill updates through bounded edits, allowing the optimizer to summarize “gradient directions” from agent experience and continuously improve procedural capability.
We evaluate SkillOpt across 6 benchmarks and 7 models, under both direct model calls and real agent execution loops with Codex + Claude Code. SkillOpt achieves best or tied-best results in 52/52 settings.
Train the skill, not the model. 🛠️🤖
🌐 https://t.co/zinqcX2wfQ
📄 https://t.co/pCI4VWdpih
Been researching agent sandbox architectures for months.
Talked to teams building on Firecracker.
Realized most of us (including me) are solving problems we don't have.
This is my first post. Spent a lot of time getting it right.
https://t.co/Yukf0p2VfH
@siddontang@sarahcat21 I agree 💯. It is the most prominent part in ai agents but there are still many.
This post shows a different perspective on agents.
https://t.co/gwCe98L0an
Been researching agent sandbox architectures for months.
Talked to teams building on Firecracker.
Realized most of us (including me) are solving problems we don't have.
This is my first post. Spent a lot of time getting it right.
https://t.co/Yukf0p2VfH
@_bgiori@Cloudflare Yup cloudflare has the primitives yet they still feel not yet.
The virtual filesystem is inspired by @Cloudflare's vibesdk but replaced with agentfs by @tursodatabase.
Been researching agent sandbox architectures for months.
Talked to teams building on Firecracker.
Realized most of us (including me) are solving problems we don't have.
This is my first post. Spent a lot of time getting it right.
https://t.co/Yukf0p2VfH
@d4rsh_tw I have been following you for some time. I like your idea of building Grind Nation to keep people motivated and grinding. You also take initiative and help others by gifting Premium and guiding them. You are always willing to help. Such a nice guy. Respect bro
Been researching agent sandbox architectures for months.
Talked to teams building on Firecracker.
Realized most of us (including me) are solving problems we don't have.
This is my first post. Spent a lot of time getting it right.
https://t.co/Yukf0p2VfH
@ayushagarwal Yup agree but both come with a cost.
Especially Sandboxes.
I dig deep and shared some unexplored paths in sandboxing.
https://t.co/gwCe98L0an
Been researching agent sandbox architectures for months.
Talked to teams building on Firecracker.
Realized most of us (including me) are solving problems we don't have.
This is my first post. Spent a lot of time getting it right.
https://t.co/Yukf0p2VfH
Been researching agent sandbox architectures for months.
Talked to teams building on Firecracker.
Realized most of us (including me) are solving problems we don't have.
This is my first post. Spent a lot of time getting it right.
https://t.co/Yukf0p2VfH
Been researching agent sandbox architectures for months.
Talked to teams building on Firecracker.
Realized most of us (including me) are solving problems we don't have.
This is my first post. Spent a lot of time getting it right.
https://t.co/Yukf0p2VfH
@threepointone The long term goal of agent sdk making it unusable at production.
I wish agents sdk works more of an extension suite and full compatible with mastra but giving more power with cloudflare primitives,so best where you should is the motto.
I am looking for an advice here.
I never built anything in software engineer for a job perspective it was all fun and eventually wanted to build one product to earn a cent trying and experimenting but failed.
Now I am at a point where I am thinking to leave all start diving deep