-If you have been talking to him for a long time and working out a long list of changes that need to be put into the main project docs, have him make a separate file with the changes that need to be implemented, including noting exactly what goes into which target doc. Then have a fresh new claude use your changes file to update the main project docs. Ideally, you want to have original claude document your changes BEFORE he compacts, so he is working from your full convo and not a summary. Using a fresh claude to update your docs eats into the weekly usage limit a lot less.

-Claude is generally going to do exactly what you tell him to do without questioning it. a lot of times, it's better to ask him how to prompt himself to do what you want, and then either sign off on or modify the plan he comes up with

-If you do a lot of work with him in a long convo and will be doing similar stuff again, ask him for a plan to optimize workflow for the same type of tasks in light of his context and usage limitations and ask him to point out what you did with him that was inefficient. He will rattle off a long list of your failures and mistakes that you never would have been aware of had you not asked.

-Almost always have extended thinking mode ON, as it definitely increases his comprehension of complex scenarios and the quality of responses. However, it does eat usage faster and the extra thinking blocks do eat into his context limits some.

(http://www.autoadmit.com/thread.php?thread_id=5841521&forum_id=2...id#49715335)