I want to get more input on one point in the pipeline AGENTS.md As it stands, the section related to prose contains a one-sentence definition and 3 short sentences of instructions. Since testers have reported some trouble with this aspect, the section could use further improvement, but I am worried that it may bloat the file too much for relatively little gain. Maxime has suggested using the deslop skill, but there doesn't seem to be a clean way to include a skill through the AGENTS file, so people would have to install it themselves. Mahesh has provided some nice rules that I could add to our file at the cost of making it a few dozen tokens longer. What do you think?

Do it. And we'll fight over it at the PR xD

Fair

'cc @Chris Hakkaart who is the master of docs-writing AI skills

We have been integrating using AI for writing and editorial reviews of our docs, and have developed a deslop skill that is working well for us. Admittedly, I have a healthy Claude subscription so have been less concerned about burning tokens. A few comments/suggestions that might be of interest: • I buried a lot of our style guide in reference files, and have our skill look them up as a part of the deslop skill. It keeps the body of the skill smaller, and can be used to selectively look up rules if they are needed. I can share these - or an adapted version can be taken/adapted from the nf-core docs already. • Using deterministic linters (cSpell and Vale) can help direct AI. In particular, Vale has a good set of rules, which can identify issues, and can work in tandem with AI. As a part of our deslop skill, Vale picks up problems, and AI fixes them. You can tune this to be more or less aggressive using Vale or AI. • Less can be more. Originally, I charged out the gate and tried to have a rule for every edge case but it did a worse job as the rules started fighting. Sometimes, trusting the model with a few clear rules is better.