Share your core rules for effective planning and builds

Discussions
1

Please suggest basic rules that can be used before any kind of builds

Someone in the forum proposed following for Plan mode


# Plan Mode Rules

1. **Avoid follow-up questions**
WHEN: Ambiguities are detected in user request/task
FIRST: try to resolve them through codebase navigation / understanding and web research.
THEN: if doubts still remain -> ASK the user follow up questions.

2. **Plan write-up and delivery**
PRE: Define a objective evaluation of confidence, as weighted measurements of:
`codebaseUnderstading` + `taskDetailQuality` + `probabilityOfScopeFulfilment` + `externalLibraryUnderstanding` - `hallucinationProbability` - `probabilityOfBugsIntroduced`
WHEN: Iterative research and codebase understanding is complete
AND: Confidence score is high
THEN: Write plan, **always** include markdown formatted `[] tasks`.
2

I use Agent Compass for all tasks.

And in prompt, I write down my expectations from the program in the Markdown format.

3

If you are planning major changes, then turn on the Agent in Agent mode and tell him that for now you will only work with him on documentation. After that, discuss the project and its changes with him.

4

Thank you for sharing Agent compass. I will check it

  1. Does it make any difference if I keep it in Agent mode and instruct to plan vs plan mode.
  2. In plan mode I can save the plan document and open it again in Agent mode later and ask to update context.
  3. That brings to another question - what is better - continue using same chat windows after letting cursor summarize the context over and over again vs save a plan, open in new chat let it use fresh context ?
5
  1. Try yourself.
  2. Try yourself.
  3. While you’re working on one task, stay in the same chat. You can also activate manual summarization by inserting /summarize at the end of the prompt, but you need to understand when there is a reason for this.
6

1 and 2 .. not sure how can I quantfy them

  1. Let me rephrase .
    Let’s say there are 3 distinct yet sequential phases of changes to be done. Now There are 2 ways to do it
    3a) Reuse the same chat window - The context , in this case, reaches 100%, then auto summarizes.
    3b) use 3 separate chat windows per phase. After phase 1 is done, initiate phase 2 with prompt to rebuild context . Thereafter continue with todo list. So with phase 3

So my question was, it there a general notion / best practise which out of 3a vs 3b is a better approach for - tokens efficiency and for quality of results ?

7

3b. And most likely you also need to commit at the end of the phase. :eyes:

8

Thank you.
And yes - I do commit at end of phase and other significant milestones.