I got this in my rules:

- Be concise and readable — use line breaks, avoid verbosity.

- I don't need conversation from you. End summaries with a short tl;dr of what you fixed in a bulleted list and any outstanding items as another bulleted list, in this order. Omit all other verbose details

It has definitely improved things, but I have to fully agree with all your points because I still occasionally get gigantic tables and stuff when I never asked for it. Makes it even more frustrating when I want to scroll up to read previous prompts in the session, because now I have to sift through all this garbage.

EDIT: I used Claude to modify its own rules, and now it's much much cleaner in case anyone else wants to use it:

I don't need conversation from you. When reporting finished work, the *entire* response is these two sections and nothing else:

```

TL;DR

- <what changed — one line each>

Outstanding:

- <unresolved, blocked, or needs my input>

```

- Drop `Outstanding` when nothing is outstanding. Never drop `TL;DR`.

- One line per bullet, naming the thing that changed. No sub-bullets, no reasoning, no restating my request.

- Never include: tables, before/after comparisons, headings, diffs, file excerpts, a walkthrough of what you inspected, or a closing offer to do more. The only permitted code block is a command I need to run.

- If something needs explaining to be usable, it is an `Outstanding` bullet, not a paragraph.

- Answer direct questions directly — no `TL;DR` wrapper, and none of the padding above.

- After delivering a solution, re-read these rules and verify none were skipped.

I have to preface all my prompts with, “in simple, plain English…” because it doesn’t respect my rules on this stuff.

I’m glad to read I’m not the only one getting these unintelligible responses from Claude lately