How much time do you spend repeating context in Claude?
by•
Curious if this is just me or other devs face it too. I use Jira and GitHub MCPs inside claude code to fetch my tickets and complete them as autonomously as possible.
But it always keeps messing up finding the correct repos, branching properly or delivering a PR without CI failing. And I end up repeating instructions and copy-pasting CI errors and asking to fix it. I tracked this and I spent on avg 75mins a day repeating context and instructions.
Anyone also facing this or found a way around it?
134 views
Replies
I've started thinking of persistent project context as almost as important as the code itself. if the agent already knows the basic rules I don't have to keep stopping and explaining the same things.
@_tallha_ijaz yeah exactly. persistence is actually the key word here. because context windows can vary, but having a fixed and portable orchestration context in place is really handy. btw, if you build stuff I have opened a solution on github https://github.com/adaOctopus/claude-plugins-app
Honestly, 75 minutes a day is pianful. I've had the same issue where I spend more time explaining the task again than actually working on it.
@_raja_tashfeen exactly, have you found a way around it ?
I've had to paste thr same error more than once too. I really wish the agent could keep that context and learn from the previous failure.
@_tina_kim hey tina, thanks for the reply and yeah i feel your frustration. context memory windows are far from great, but in my opinion they are not even the biggest part of the problem. but instead orchestration is. because even when claude has the context does not always know what to do with it. in case you are interested i have opened a solution on github - https://github.com/adaOctopus/claude-plugins-app if you want to try it out and let me know
things like the right repo, branch rules , test command and CI steps are worth writing down once instead od explaining them in every chat.
@_daniel_smith exactly. i do believe that it works well when they are portable and fixed. if you are a developer you can use my solution for free - https://github.com/adaOctopus/claude-plugins-app
I have experienced the same agents making small workflow mistakes. I started documenting commands, repo structure and deployment rules before asking the agent to work.
Same shape of problem here, but the fix that actually held for me was not a bigger context file, it was making the rules unskippable rather than just available. A model reads a 1500 line reference doc and still sometimes glides past the one line that matters because reading it is optional. What stuck was turning the recurring corrections into checks the agent has to state pass or fail on before it is allowed to act, not text it might glance at.
Concretely: our repeat failure was not wrong code, it was right code posted to the wrong environment or the wrong account. Writing that down as a paragraph did nothing, it happened again three sessions later. Making it a numbered gate the agent has to clear before the write step is what actually stopped it.
Worth splitting your 75 minutes into which kind you're repeating. If it's facts, repo names, branch rules, a file solves it once. If it's discipline, don't skip step 3 even when step 3 feels obvious, a file won't hold. You need something the agent is forced through, not just shown.
claude made failing test pass by weakening assertion instead of fixing code, suite stayed green for days before I noticed ) now have to paste exact same instruction into evry new chat forbidding it to touch test assertions
Do you create a project and define the context and project instruction? Because that's something I do and I have not encountered what you're describing.
@richatsealedvault yeah yeah i do. context engineering is always there in my projects. however i realize sometimes is not enough. like for example i use Jira and Github mcps to fetch context, but everytime i was asking to execute a task it was messing up things that were related to repos and acceptance criteria.
So you spend 75 minutes a day telling others what to do. Welcome to the new way of coding :D
I think i spend more time fine tuning each agent to do what i want it to do, but then again, i don't have to spend 20 hours doing that work.
@cryptosymposium quite a little these days. I keep the recurring rules and repo details in Notion rather than a chat thread, so there's one fixed place the agent checks instead of me retyping the same instructions every session. What still gets missed is usually the thing I never wrote down because I assumed it was obvious. Worth checking whether your 75 minutes is mostly context you haven't documented yet, or context you have documented and the agent still skips past anyway. Those need different fixes.