Claude code max plan is exhausted every time agent start working

[akshaynikhare]

Can someone suggest optimisation strategies to reduce the usage of context window ,

And how I save my token consumption

I have max plan I feel in around 10 min of running paperclip

My 5 agent just barely run heard best and 1-2 take and the limit is exhausted

I was running. It for 3 day and my weekly limit is also gone now

What should I do

[VivaLaMikey]

I had this same issue, it seems that it's very token hungry.

[Shieldxx]

What model? I'm on the 20$ plan and Sonnet is ok for CEO but I use Codex or lmstudio for everything else because of token usage

[akshaynikhare]

i use max plan. , CEO , CTO on Sonet. and for developer i use OPUS , i also have lmstudio. but its too slow

[Rangozhang]

Me too. I think it needs a session rewind system. Even it's only me replying under an issue, the agent will start from:

Let me check my inbox for any new assignments.
EXECUTED COMMAND

Interesting — XXX-31 is back to todo status. I marked it done in my previous heartbeat, but the status shows todo now. This might be because the routine has reset it. Let me check the heartbeat context to see what happened.

Actually, looking at the activeRun.invocationSource: "automation" and triggerDetail: "system", this might be a new automated heartbeat for the same daily review task. The task was reset to todo automatically.

Let me check the current issue state.
WORA-31 is back to todo — looks like it was reset by automation. Let me check the current state.
EXECUTED 2 COMMANDS

I can see that the latest comment is the review I posted in the previous heartbeat (created at 22:33:10, while the new run started at 22:53:56). The task was reset to todo at 22:53:56 (by automation?). There's a comment at 22:53:56 that I haven't seen yet. Let me check all 3 comments. I see totalComments: 3 and the latest comment ID is 597b2672-2ea3-4d35-a2d2-3fc6098a041b - but the comment I just read has ID 5e9c7d56-ff2e-44bd-b33c-94f5c6bc0865. So there's a newer comment I haven't seen.

I need to pull that latest comment to see what it says.
There's a newer comment (id 597b2672) I haven't seen. Let me read it.

[brendonpeter]

Have you changed the default Agent.md file at all? After burning through all my max credits I realised changing the Agent file breaks the entire agent workflow/entire platform logic

[akshaynikhare]

@brendonpeter yes I also have that issue , when shortened the agent.md to much in the middle of the session it affect the in progress task