bluebag.ai

We’re building a better way to ship reliable AI agents for the web

[bluebag.ai @BluebagAI]

Introducing Bluebag.

World's first Agent Skills Platform for Typescript.

You'll never write AI-SDK and Langchain agents the same again ⬇️

[bluebag.ai @BluebagAI]

@akbuilds_ This has a cool score of 100

[bluebag.ai @BluebagAI]

@tarunsachdeva Very clean UX, love it

[bluebag.ai @BluebagAI]

Load any Public Skill.

In your ai-sdk or Langchain agents at runtime.

[bluebag.ai @BluebagAI]

@CameronMThacker @vercel IMO, I think the approach works great for scoped problem sets, e.g., pass on to a subagent.

Scaling is a valid question.

My thoughts are that the main problem is "untriggered Skills". Fix that, and you have a better comparison

https://t.co/vHfd8L1Y8Z

[bluebag.ai @BluebagAI]

A bit of an apples-to-oranges comparison to compare untriggered Skills vs compressed Agents(dot)md file

If the Skills aren't triggered, they don't exist (to the model)

Nothing to compare (in terms of execution)

The sound comparison would be to fix the Skills trigger, then compare deterministic executions across both

And in the future, the tooling around Skills needs evals. Evals for Skill triggers

[bluebag.ai @BluebagAI]

@intellectronica @trq212 @sughanthans1 100%. For engineers, the “aha” lands even harder when Skills are remote + composable (not trapped on one's machine).

Demo idea: load a Skill into a Vercel AI SDK / LangChain (TS) agent and run it from anywhere.

[bluebag.ai @BluebagAI]

@gary_qz @vercel It comes to triggering a Skill. Same way you have evals for good ol' prompts.

If your Skills aren't triggered, they effectively don't exist.

A better comparison would be to fix the Skill triggers (use evals or other means), then compare against a deterministic workflow

[bluebag.ai @BluebagAI]

Nicely done.

nitpick: In the Creating Hedera Agent Kit Plugins Skill, you'll get better results by writing out the exact scripts.

Rather than "Install dependencies: Set up a TypeScript project with hedera-agent-kit and @hashgraph/sdk"

Basically, the scripts from https://t.co/SBBoN5okqL

[bluebag.ai @BluebagAI]

@Joseph1149010 @doodlestein Very good read, but seems like a wrong comparison

https://t.co/s0KxST87qW

[bluebag.ai @BluebagAI]

A bit of an apples-to-oranges comparison to compare untriggered Skills vs compressed Agents(dot)md file

If the Skills aren't triggered, they don't exist (to the model)

Nothing to compare (in terms of execution)

The sound comparison would be to fix the Skills trigger, then compare deterministic executions across both

And in the future, the tooling around Skills needs evals. Evals for Skill triggers

[bluebag.ai @BluebagAI]

@0xcgn @steipete One of the best rules I've found is to add something like:

"When in doubt, ask questions and await my approval before continuing. Do not assume. Simply ask me, and I'll confirm what to do"

Helps with brainstorming before rushing off to confidently create wrong solutions

[bluebag.ai @BluebagAI]

@cursor_ai Would’ve been great to include a quick example of how these traces would render in common observability tooling (e.g., spans/attributes in OTEL, a sample timeline in a trace UI).

The spec is “UI agnostic,” but a reference visualisation would help adoption.

[bluebag.ai @BluebagAI]

A bit of an apples-to-oranges comparison to compare untriggered Skills vs compressed Agents(dot)md file

If the Skills aren't triggered, they don't exist (to the model)

Nothing to compare (in terms of execution)

The sound comparison would be to fix the Skills trigger, then compare deterministic executions across both

And in the future, the tooling around Skills needs evals. Evals for Skill triggers

[Vercel @vercel]

We're experimenting with ways to keep AI agents in sync with the exact framework versions in your projects. Skills, 𝙲𝙻𝙰𝚄𝙳𝙴.𝚖𝚍, and more.

But one approach scored 100% on our Next.js evals:
https://t.co/8ACw9BgudB

[bluebag.ai @BluebagAI]

@bhaidar Appreciate you sharing it, Bilal 🙏

What are you building with the AI SDK right now (agent, some workflow, or automation)?

[Bilal Haidar @bhaidar]

🔥 This is exactly what I needed!

Vercel AI SDK + Agent Skills = game changer for building smarter AI agents 🚀

Perfect for anyone building:
- Custom chatbots
- AI automation tools
- Specialized agent workflows

The thread breakdown is 🔥 - check it out 👇

#AI #Vercel #AgenticAI #WebDev

[bluebag.ai @BluebagAI]

@itplaysout Indeed. Give it a try!

[bluebag.ai @BluebagAI]

Building Agents with Vercel's AI-SDK?

Here's how to use Agent Skills in your AI agents:

(1/5)

[bluebag.ai @BluebagAI]

Interesting proposal.

Also fits the potential model of documentation sites as Skills.

[Matt Silverlock 🐀 @elithrar]

We're proposing an approach to make discovering agent skills easier by using the .well-known URI standard + a index of associated files for agents to pull down.

There's tools (like add-skill) to add skills, but you have to find them first.

RFC: https://t.co/9Qrf0aXJ5D

[bluebag.ai @BluebagAI]

Love this thread, follow Bluebag for more

Check out the docs for real-world examples and advanced production usecases

(5/5)

[bluebag.ai @BluebagAI]

But what happens under the hood?

Your agents immediately:

- Get access to runtime virtual machines
- Run complex scripts and read Skills on-demand
- Leverage a rich file API to mint download links

All without managing any infrastructure

(4/5)