@trq212: a prompt I've been using a lot recently: implement <SPEC> and while you do, keep a running implementation-notes.html fi…

as much as you spec there are always still ambiguities and unknown unknowns that come up and this gives the model a good out to make decisions but keep you in the loop

okay this is going kinda viral and tbh my original text was kind of messy, so here's a second pass with the help of Claude: -- Implement <SPEC>. As you work maintain a running implementation-notes.html file that captures anything I should know about how the implementation

This should not be the way in 2026

why?

Mine was DECISIONS.md but this is a nice evolution of that. My workflows still require that I make key decisions, but tracking them helps keep Claude aligned with my vision.

This reminds me of developer diaries. Unfortunately I was always too lazy to write down decisions and findings. Fortunately, Claude does a great job in documenting and with a developer diary skill this burden is taken away from me: https://github.com/ikangai/claude-skills/tree/main/diary…

I created a Human Agent Context (HAC) layer skill for this: https://github.com/pytholic/claude-skills/tree/main/hac-init…

cool, i follow the same But i say note down the critical queries need decision instead of self decisions and i've shifted to HTML in recent days, it was amazing productive shift

been doing a version of this, asking Claude to keep a changelog.md as it goes. the magic is that next session you paste it back and skip the 20 min of "why did I structure it this way." the notes are worth more than the code sometimes.

Good but an upfront grill-me step is still essential to minimize the number of decisions the AI will otherwise make without input. Of course no solution can be specified completely, unknown unknowns will come up. But the solution is always better when I have a say in it.

This right here explains everything about the quality issues in Claude Code. You should know better than this, sir.

That loop is the interesting bit: docs stop being an artifact and become shared working memory. The trick is receipts: code links, diff history, and a clear owner for conflicts when agent and human disagree.

The most useful one I used most frequently is 「Every word should serve a purpose.」, give it a try，you'll come back and thank me later.

same, kinda similar to the instructions I am using where I ask the LLM to write a md report in docs/reports/ after completing a task. especially when doing sensitive tasks in prod env, etc...

Just implementation-notes.md instead of .html. the md viewer of VSC is quite capable of rendering and saving tokens on html.