3 ms·

Show HN: AutoLearn Skills for self-improving agents

27 points by toobulkeh 1y ago

personjerry 1y ago
I don't understand, it'd be nice to see even a single example of the application
- toobulkeh 1y ago
  Demo video: https://youtu.be/s_9m4P9_6jc?si=P7By-zl62GkFMRpH https://youtu.be/s_9m4P9_6jc?si=P7By-zl62GkFMRpH
toobulkeh 1y ago
I vibe coded this at a hackathon a few weeks ago. The solution isn’t stable yet, but it’s similar to Claude Skills with the benefit of only needing an MCP client. I open sourced it too. It has a bit to go—like automatically monitoring reasoning traces and security—but I thought this would be easier to use than trying to maintain your own skills.md collection if the server handled more of it for you. I’d be curious if anyone thinks I should keep working on it. I’d love someone to collaborate with on it!
- diamondfist25 1y ago
  It is a novel idea and one that I am interested in. Would you mind sharing your repo, would love to check it out
  - toobulkeh 1y ago
    Git repo: https://github.com/tarkaai/autolearn https://github.com/tarkaai/autolearn It’s like 5 hours of work and I’m super early with it!
    - itsajchan 1y ago
      I hosted this Hack Day up in Seattle at the GitHub office and this project was one of 10 from an event that brought in nearly 60 project submissions! @toobulkeh is an awesome builder and was a pleasure to see this project come to life in a few short hours. Come check AutoLearn and all the other projects here! https://hackersquad.io/showcase/events/cmf08dkr00004ph0kdzxphqvv https://hackersquad.io/showcase/events/cmf08dkr00004ph0kdzxp...
drdeca 1y ago
How does it determine whether the code it has for the skill, failed? (In order to determine that it needs to improve the code for that skill, I mean)
- toobulkeh 1y ago
  Definitely needs some work here. That isn’t done (yet). Great feature request! I think it would be some kind of testing logic or feedback loop. Or both!
stpedgwdgfhgdd 1y ago
I sometimes have to decide between running Claude Code headless with a prompt or let CC generate an application based on the prompt. (coining: Codification Threshold tradeoff) The tradeoff is that the prompt is non-deterministic, but also more flexible. It can handle unknown situations. The generated code executes much faster and reliable for known situations. If auto-learn would reduce the non-deterministic nature, i.e. chance of failing, that would speed up software development.
snowfield 1y ago
This is a very interesting concept. I have a few questions though. Even if the agent has mcp tools avaible, you can't guarantee that it would necessarily use them right? I imagine that you'd have a lot of new tools that do nearly the same things. Also how does this scale? If the agent has like 300 tools.
- toobulkeh 1y ago
  I don’t know (yet) but some ideas might be: 1. A skill that searches skills 2. MCP that returns only relevant skills based on context 3. Separate concerns through sub agents that each have their own autolearn server

27 points by toobulkeh 1y ago

personjerry 1y ago

I don't understand, it'd be nice to see even a single example of the application

toobulkeh 1y ago
Demo video: https://youtu.be/s_9m4P9_6jc?si=P7By-zl62GkFMRpH https://youtu.be/s_9m4P9_6jc?si=P7By-zl62GkFMRpH

toobulkeh 1y ago

I vibe coded this at a hackathon a few weeks ago. The solution isn’t stable yet, but it’s similar to Claude Skills with the benefit of only needing an MCP client. I open sourced it too. It has a bit to go—like automatically monitoring reasoning traces and security—but I thought this would be easier to use than trying to maintain your own skills.md collection if the server handled more of it for you. I’d be curious if anyone thinks I should keep working on it. I’d love someone to collaborate with on it!

diamondfist25 1y ago
It is a novel idea and one that I am interested in. Would you mind sharing your repo, would love to check it out
- toobulkeh 1y ago
  Git repo: https://github.com/tarkaai/autolearn https://github.com/tarkaai/autolearn It’s like 5 hours of work and I’m super early with it!
  - itsajchan 1y ago
    I hosted this Hack Day up in Seattle at the GitHub office and this project was one of 10 from an event that brought in nearly 60 project submissions! @toobulkeh is an awesome builder and was a pleasure to see this project come to life in a few short hours. Come check AutoLearn and all the other projects here! https://hackersquad.io/showcase/events/cmf08dkr00004ph0kdzxphqvv https://hackersquad.io/showcase/events/cmf08dkr00004ph0kdzxp...

drdeca 1y ago

How does it determine whether the code it has for the skill, failed? (In order to determine that it needs to improve the code for that skill, I mean)

toobulkeh 1y ago
Definitely needs some work here. That isn’t done (yet). Great feature request! I think it would be some kind of testing logic or feedback loop. Or both!

stpedgwdgfhgdd 1y ago

I sometimes have to decide between running Claude Code headless with a prompt or let CC generate an application based on the prompt. (coining: Codification Threshold tradeoff) The tradeoff is that the prompt is non-deterministic, but also more flexible. It can handle unknown situations. The generated code executes much faster and reliable for known situations. If auto-learn would reduce the non-deterministic nature, i.e. chance of failing, that would speed up software development.

snowfield 1y ago

This is a very interesting concept. I have a few questions though. Even if the agent has mcp tools avaible, you can't guarantee that it would necessarily use them right? I imagine that you'd have a lot of new tools that do nearly the same things. Also how does this scale? If the agent has like 300 tools.

toobulkeh 1y ago
I don’t know (yet) but some ideas might be: 1. A skill that searches skills 2. MCP that returns only relevant skills based on context 3. Separate concerns through sub agents that each have their own autolearn server