DEV Community

Cover image for AI Agents Still Can’t Use Your Stack. I Built a Fix.
Harikrishnan
Harikrishnan

Posted on

AI Agents Still Can’t Use Your Stack. I Built a Fix.

We’ve made documentation readable for machines.

We’ve built agents that can run workflows, call tools, and execute multi-step logic.

And still… they don’t really work with the stacks we actually use.

If you’ve ever gone looking for an existing Agent Skill for something even slightly niche, you know the feeling. You search. You scroll. Nothing fits.

Your stack exists.

Your docs exist.

Your agent exists.

But they don’t speak the same language.

That’s the friction.

The Missing Layer

We now have llms.txt. It gives documentation a structure that models can read without guessing.

We also have Agent Skills. These let agents do real work by loading instructions, workflows, and domain-specific logic when needed.

But they solve different problems.

  • llms.txt helps agents understand systems
  • Agent Skills help agents operate them

Right now, moving from one to the other is still manual work.

Even if your documentation is perfectly structured, someone still has to sit down and turn it into a skill.

That’s the part that doesn’t scale.

So I Built: txtskills

txtskills takes documentation and turns it into usable agent skills.

You drop in an llms.txt URL (or even just a docs base URL), and it:

  • Pulls the documentation
  • Interprets the structure
  • Converts it into an installable Agent Skill

From there, the generated skill can be installed into any compatible coding agent with a single command.

Claude Code, Amp, Antigravity, VS Code, or any environment that supports the open Agent Skills format.

No rewriting workflows by hand.

No packaging instructions manually.

Just:

Docs → Understanding → Execution

Why This Matters

Agent Skills are how teams are starting to capture real working knowledge.

Not just prompts, but:

  • internal processes
  • repeatable workflows
  • operational logic

Once written, these can be reused across agents.

The problem is getting them built in the first place still takes time.

txtskills handles that conversion step.

If documentation exists, turning it into something an agent can actually use shouldn’t be a separate project.

Try It

Live: https://txtskills.hari.works/

GitHub: https://github.com/hk-vk/txtskills

Upvote on Peerlist: https://peerlist.io/harikrishnanvk/project/txtskills--convert-llmstxt-to-agent-skills

Top comments (0)