Hello everybody, I am getting berserk writing SKILLS.md for everything I work on. Where would be the best place in an nf-core pipeline for a skills file instructing the agent what can be done with the output of the pipeline?

@Igor Trujnara is the person to ask. We're currently working on a standardised solution for these kinds of questions. See github.com/nf-core/agents

So far, we are quite advanced with agent documentation for creating pipelines (AGENTS files plus modules skills), but we have not worked in the direction of agents using pipelines. That said, released pipelines should come packaged with solid documentation on running the pipeline (

usage.md

) and the outputs (

output.md

). An agent should be able to figure out what it can do with the files, and if not, it is a good reason to contribute to the standard human documentation. We could consider having a general "running nf-core" skill, but we have not had this discussion so far.

Thanks for the answers, yes it is an agents file for using the pipeline and planning the next steps with its output. It surely has a lot of overlaps with

output.md

but has a couple of important differences: • It does not need to be illustrated, • It can cover multiple versions of the pipeline, • It does not have to avoid redundancy with

usage.md

or

README.md

. • It can get in more details in the syntax of the tools used in the modules, when it is expected that some users may want to re-run just a command without running the whole pipeline again, • Likewise, it can get in more details about the downstream tools. Obviously a nf-core pipeline is small enough that an agent can read and analyse it entirely quite fast, but I am designing the skill file for me and my colleagues who interact regularly with the pipeline's output in various contexts. I am still figuring out if it is useful 🙂

I don't think it should cover multiple versions of the pipeline - it's committed to a specific version if it's in the repo. I think it should only correspond to that version if it's hosted there (fine to reference old versions but not be designed to work across all versions)

FYI, there is an agent for the #C0860UN5GQ5 pipeline: github.com/Sheng-T/methylongAgent

Nice! Though that's more for running the pipeline it seems? Rather than downstream analysis.

Yeah, I think it is intended to lower the barrier of entry, I think the idea is to deploy it on a server to have an interface for normal people

normal people 😆

normal people?