I use openclaw a lot, and I spent a good amount of time tweaking it to get it how I want it. A good part of this is creating skills that allow it to do things I need. While doing this, I have found many times it wasn't using the skills I want it to use. This happens a lot when setting up new skills and trying to get the description and trigger phrases ironed out so it uses the skill naturally.
I found myself asking my agents "What skills did you use?" all the time and it was getting annoying. With this line added to your TOOLS.md, you will never have to do this again.
- Always Show Skills Used: When using any skill, include
Skills Used:at the end of your response listing which skill(s) were used.
Now I never have to ask, and I also can see when it finds creative usages to the skills I make available.
Another problem I had that was solved in a similar way, is when files are edited and I wasn't aware of it.
- Always Show Files Modified: When editing or creating files, always show files modified at the end. Files
Files Modified:.
This simple addition to your TOOLS.md and you never have to guess!
I am impressed at how well my agents do this without being reminded. This is really handy when troubleshooting skills execution.
This is also very handy in harnesses and is not limited to openclaw.
https://bsky.app/profile/did:plc:2zx7of5ioe5hvc4suhlvnguo/post/3mh5yogqzdc2l
https://bsky.app/profile/did:plc:2zx7of5ioe5hvc4suhlvnguo/post/3mh5yogqzdc2l
The rewards earned on this comment will go to the author of the blog post.
Interesting. I still don't quite understand it all, but I think I am starting to.
Setting up Open Claw just seems too risky security wise for me, I just use Chat GPT Codex...
you can set it up on a $4/month VPS to check it out. worth the price...
It’s as risky as you allow it to be.
Dona eh....
From suits. Gilfoyle from Silicon Vally and Sherlock from Scotland Yard.
Avengers :) .... with Tony Starck behind the desk
I do plan to have more personalities, but the job is the first priority, then I come up with the personality to go with it.
I did something similar, since I'm using different models for different things, I did the same with one of my agents, to always say what model was used for that task.
Not a bad idea if you swap models a lot.
not bad for research!
Great idea. Thank you for sharing. Will try this out as well.