You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@magpie.apache.org by Jarek Potiuk <ja...@potiuk.com> on 2026/08/17 23:23:26 UTC
[PROPOSAL] Getting more "human" input for our SKILLs
Hello here,
I am definitely now going to focus back a lot of my time on Magpie (for
reasons ;) other stuff was taking more priority). Besides attempting to
lead several open threads to completion - and trying to resolve the
"installation" and "permissions" difficulties, I thought of one thing where
we all could be helpful.
Most of the SKILLs and instructions of ours are mostly AI generated - which
makes those skills less efficient than they could be.
I looked at other SKILLs - for example Claude Code Security (which works
very well and very predictably) - and many of those skills are much clearer
- visibly written by humans.
I thought it might be a good opportunity - before next release to split
the current - important - SKILLS we have and apply some human content there
- basically simplifying and rewriting those in "Human English" - simpler.
shorter, and also ensuring they don't exceed the context. I think this
could result in a much better baseline; even future generated changes
should improve as they follow this standard.
It might also give an opportunity for those who haven't closely examined
what Mapgie can do to try some of the things and understand more how those
particular skills work, use and rewrite them to test in their projects.
I think solving the installation and permissions issues will address "how
easy it is to start and use," while this effort might help with "how well
it works."
I'm raising this here: We could automatically create a small issue where we
could all sign up for a particular part of Magpie to try and get the "feel
of it" while improving it - if that seems like a good idea.
Side comment - today I saved likely DAYS of effort on importing some
**redacted** results of **redacted** pile of **redacted*" analysing those,
assessing those and assigning it for further detailed look. It works.
J.
Re: [PROPOSAL] Getting more "human" input for our SKILLs
Posted by Jarek Potiuk <ja...@potiuk.com>.
Yep. That was exactly what I was thinking.
On Tue, Aug 18, 2026 at 5:26 PM Yeonguk Choo <ch...@apache.org> wrote:
> Sounds like a great idea!
>
> Rather than just rewriting the skills in “Human English,” I think it would
> be more valuable to actually use them first, read through the skills while
> using them, and then improve them based on what we learn from that
> experience.
>
> That way, we can not only make the skills clearer and shorter, but also
> better understand how they actually work in practice.
>
> Thanks,
> Yeonguk
>
> On 2026/08/17 23:23:26 Jarek Potiuk wrote:
> > Hello here,
> >
> > I am definitely now going to focus back a lot of my time on Magpie (for
> > reasons ;) other stuff was taking more priority). Besides attempting to
> > lead several open threads to completion - and trying to resolve the
> > "installation" and "permissions" difficulties, I thought of one thing
> where
> > we all could be helpful.
> >
> > Most of the SKILLs and instructions of ours are mostly AI generated -
> which
> > makes those skills less efficient than they could be.
> >
> > I looked at other SKILLs - for example Claude Code Security (which works
> > very well and very predictably) - and many of those skills are much
> clearer
> > - visibly written by humans.
> >
> > I thought it might be a good opportunity - before next release to split
> > the current - important - SKILLS we have and apply some human content
> there
> > - basically simplifying and rewriting those in "Human English" - simpler.
> > shorter, and also ensuring they don't exceed the context. I think this
> > could result in a much better baseline; even future generated changes
> > should improve as they follow this standard.
> >
> > It might also give an opportunity for those who haven't closely examined
> > what Mapgie can do to try some of the things and understand more how
> those
> > particular skills work, use and rewrite them to test in their projects.
> >
> > I think solving the installation and permissions issues will address "how
> > easy it is to start and use," while this effort might help with "how well
> > it works."
> >
> > I'm raising this here: We could automatically create a small issue where
> we
> > could all sign up for a particular part of Magpie to try and get the
> "feel
> > of it" while improving it - if that seems like a good idea.
> >
> > Side comment - today I saved likely DAYS of effort on importing some
> > **redacted** results of **redacted** pile of **redacted*" analysing
> those,
> > assessing those and assigning it for further detailed look. It works.
> >
> > J.
> >
>
Re: [PROPOSAL] Getting more "human" input for our SKILLs
Posted by Yeonguk Choo <ch...@apache.org>.
Sounds like a great idea!
Rather than just rewriting the skills in “Human English,” I think it would be more valuable to actually use them first, read through the skills while using them, and then improve them based on what we learn from that experience.
That way, we can not only make the skills clearer and shorter, but also better understand how they actually work in practice.
Thanks,
Yeonguk
On 2026/08/17 23:23:26 Jarek Potiuk wrote:
> Hello here,
>
> I am definitely now going to focus back a lot of my time on Magpie (for
> reasons ;) other stuff was taking more priority). Besides attempting to
> lead several open threads to completion - and trying to resolve the
> "installation" and "permissions" difficulties, I thought of one thing where
> we all could be helpful.
>
> Most of the SKILLs and instructions of ours are mostly AI generated - which
> makes those skills less efficient than they could be.
>
> I looked at other SKILLs - for example Claude Code Security (which works
> very well and very predictably) - and many of those skills are much clearer
> - visibly written by humans.
>
> I thought it might be a good opportunity - before next release to split
> the current - important - SKILLS we have and apply some human content there
> - basically simplifying and rewriting those in "Human English" - simpler.
> shorter, and also ensuring they don't exceed the context. I think this
> could result in a much better baseline; even future generated changes
> should improve as they follow this standard.
>
> It might also give an opportunity for those who haven't closely examined
> what Mapgie can do to try some of the things and understand more how those
> particular skills work, use and rewrite them to test in their projects.
>
> I think solving the installation and permissions issues will address "how
> easy it is to start and use," while this effort might help with "how well
> it works."
>
> I'm raising this here: We could automatically create a small issue where we
> could all sign up for a particular part of Magpie to try and get the "feel
> of it" while improving it - if that seems like a good idea.
>
> Side comment - today I saved likely DAYS of effort on importing some
> **redacted** results of **redacted** pile of **redacted*" analysing those,
> assessing those and assigning it for further detailed look. It works.
>
> J.
>