Ask HN: How did you improve agent's output readability?
It's been a couple of months now that I'm seeing tons of complaints about Claude's output. It is VERY difficult to read and makes reasoning particularly tedious. Asking it to use ASD-STE100 improved things a bit but it's still not perfect.
I've also seen some tricks on Twitter such as adding this kind of blocks in the AGENTS.md/CLAUDE.md file:
---
- Use clear subject/verb/object constructions. Do not use cleft sentences, contrastive appositives, appended-glosses, or trailing clauses.
- Assume I may edit documents myself. Especially markdown documents.
- When writing markdown documents, don't include references to conversations or threads a reader would not know about.
---
I'm curious how you're handling this. Did you find anything that improves an agent's communication style? Also, do we know why it's happening?