Limit Skills and Hooks to individual agents only

[zikajk]

To upvote this issue, give it a thumbs up. See this list for the most upvoted issues.

The former isn't possible yet; the latter can be handled by the hook's code itself, but it would probably make sense to allow doing it via configuration as well. A few people have already asked me for this, so I don't want to forget about it again 😅.

For skills, it would likely just be a new field in the frontmatter. For hooks, it would either be a new matcher property (if the current design allows this without breaking), or the hook would be created directly under the agent's configuration...

[ericdallo]

Interesting, I think it makes sense, the same did here for variants right?
the hook is not clear as well to me, but maybe a simple regex or matcher would be enough

[zikajk]

I’m not sure what the best approach for those hooks would be. We could add a new field to the hook definition in the config, but then the question is whether it makes sense for everything else to be defined inside the agents themselves, for example in Markdown, while hooks suddenly reference an agent.

Of course, the hook script can already detect which agent is currently running, but that requires more work than handling it through configuration. I’ll try to ask what use cases they needed it for, or see whether another harness already solves this in a way we could take inspiration from. There’s no rush.

And I think you’re probably right about skills. I can imagine defining them either as a list in the agent’s Markdown file or in JSON, with only the enabled skills being listed, perhaps using regexes too. Something like allowed-skills.