Let me / multiple skills in one message

[roblourens]

No reason this shouldn't be allowed, right? I want to tell the agent to do something and reference multiple skills explicitly

[aeschli]

If it's about referencing wouldn't that be a context (with hash)?

[roblourens]

We usually use / with skills right?

[aeschli]

The slash indicates a command. Multiple slash commands in a line would be multiple commands, likely executed in order. We could do that but it would probably mean to also support this for built-in commands (/clear), contributed commands (/plan), prompts files (that can potentially change model/tools)...

[roblourens]

I was talking to Paul about how I use skills for everything, and I use them in a sort of loose fuzzy way. I don't want a /skill to mean "deterministically execute the skill right now", because I have an LLM and it can decide what to do. The Codex app lets me / multiple skills so I can just talk to it like a human. Examples of how I've used skills:

/create-skill based on /foo and /bar, except do xyz instead

/plan this, then merge origin/main again, then /implement

Verify this feature, use /code-oss-debug or /code-oss-logs if you get stuck

[aeschli]

Agree that looks smooth, but I don't see yet how this fits in our world where some slash commands are implemented on the UI side, and some map to prompt files that can change tools and models.
What you really do is attaching skill:foo and skill:bar as context and IMO the proper way would be to use #skill:foo for that (which doesn't exist yet). Maybe we can allow the slash completions and just treat it like #skill:foo