I am loving this process for skill making:
- setup subagents that can do deep research - ask for X research runs covering different angles of a thing - distill the research reports into a single SKILL.md - include research alongside skill for reference
Many users affirm the outlined AI subagent workflow for creating skills because it matches their existing loops and stresses synthesis over mere breadth.
No Digg Deeper questions have been answered for this story yet.
If the skill needs a tool, you can also ask for a dependency-free cli to go with it.
I am loving this process for skill making:
- setup subagents that can do deep research - ask for X research runs covering different angles of a thing - distill the research reports into a single SKILL.md - include research alongside skill for reference
@filicroval I've been using Gemini's deep research agent a lot, but it can be a bit pricey: https://ai.google.dev/gemini-api/docs/deep-research
I want different angles, so it also works well to ask different models. Usually that's a skill plus web search
@fofrAI interesting approach. do u version the SKILL.md separately or just let the latest distill override?
@fofrAI what's your go-to model for deep research?
@fofrAI Is it possible to get the same effect without using skills.md, just a simple call to llm or tool call?
@fofrAI this is exactly my loop now. the distill step is where it lives or dies for me.
@fofrAI oh i see, i never tried, i think i'll have a look at that
@fofrAI I have this as a skill which I call meta-skill
@fofrAI Tks for sharing, how you orchestrate gemini deep research agent? Would it handoff to other on-device agent too
@fofrAI agents cover the angles, you synthesize. that's the move. hard part isn't breadth, it's deciding what actually matters.