What to avoid
Seven habits that use up your Claude limit, and what to do instead
Claude's limit refills every five hours and every week. Some habits drain it much faster than others, even when the work gets done. The examples below come from a founder's build session: grading fixes, skill updates and documentation.
Three of these showed up in your report today. Start with #2, your biggest share, then #1 and #5.
1
Avoid: Keeping one chat going all dayIn your report today
Why it matters: Every time you send a message, Claude rereads the whole conversation from the top. Caching makes the rereading cheaper, but message 40 still rereads about ten times as much as message 4.
Do this instead: Start a new chat for each new task. If you need earlier context, paste a two-line summary.
In your work: Do grading rounds in a fresh chat with a two-line hand-off instead of extending one build session past 150 steps.
2
Avoid: Dropping whole files into the chatIn your report today
Why it matters: Claude reads every page or row, then rereads it with every message that follows.
Do this instead: Paste only the part you need, or ask Claude to work on the file in its folder and give you just the answer.
In your work: Batch all report fixes, then render and check the report once instead of after every round.
3
Avoid: Leaving connected apps on that you don't useMedium
Why it matters: Each connected app (email, calendar, a CRM) brings its own instructions into every chat, and a connection that pulls full records reads far more than you need.
Do this instead: Turn off connections you don't use for this work, and ask for only the fields you need.
In your work: Leave the browser and email connections off while building skills unless the work needs them.
4
Avoid: Asking for long write-ups when a short answer will doMedium
Why it matters: What Claude writes back is the most expensive part of each exchange.
Do this instead: Ask for the shape you want: a table, three bullets, or just the email.
In your work: Ask for a one-line summary of what changed, and request the full skill text only for the final review card.
5
Avoid: Using the most powerful model for routine workIn your report today
Why it matters: Bigger models use the limit faster. For routine work they rarely give a better result.
Do this instead: Use the standard model for sorting, summarizing, reformatting and routine drafts. Switch up for hard judgment calls where a mistake is costly.
In your work: Rebuilding, zipping and saving copies are routine. Keep the bigger model for designing the skill's logic.
6
Avoid: Re-pasting the same reference material every dayMedium
Why it matters: Pasted material is read fresh each time. Material kept in a Claude Project can be cached when it is reused, which counts somewhat less against the limit while the cache is fresh.
Do this instead: Put documents you use over and over in a Project once.
In your work: Keep the ShowEm document style and house voice in a Project instead of loading them fresh each session.
7
Avoid: Piling up instructions and skills you no longer useGrows with steps taken
Why it matters: Standing instructions, memory and the skill list load at the start of every chat and are re-read with every step Claude takes, so the more steps a task takes, the more they add up. A fresh chat does not shrink this.
Do this instead: Keep instructions short and current, turn off skills and connections you don't use, and ask for related changes together so Claude takes fewer steps.
In your work: Turn off skills you haven't used in a month. The skill list is re-read with every step.