AI Lab · by Solar Orbit · 2026-10-20

Claude Code now warns you before an old chat eats into your 5-hour limit

Small but practical change: Claude Code now warns when resuming a long old conversation will burn a big chunk of your usage window. Long contexts are expensive. How do you manage context hygiene with coding agents?

38 sparks · 14 comments

Comments

Noah Campbell · 14 likes

compaction helps but you lose details. tradeoffs everywhere

Brian Otieno · 6 likes

yeah compaction sometimes forgets the one weird constraint you told it

Isha Menon · 11 likes

new task = new session. i treat it like closing browser tabs

Avery Johnson · 3 likes

context rot is real

Sara Ahmed · 0 likes

this is the way. old context just confuses it anyway

Narin Chai · 7 likes

learned this the hard way when one 'quick question' in an old chat ate half my limit

Midnight Mango · 7 likes

the 'quick question' in a 200k context thread is never quick 😂

Mia Bauer · 5 likes

my 5 hour limit lasts about 45 minutes on a good day

Sleepy Lantern · 3 likes

you might be the problem 😅

Gentle Baker · 1 likes

power user problems

Rafael Almeida · 0 likes

I keep a NOTES.md in the repo the agent updates, then start fresh sessions that read it

Olena Shevchenko · 0 likes

poor man's memory, works great

Ethan Brooks · 0 likes

good UX change honestly, surprises in billing are the worst

Quiet Fern · 9 likes

warnings > surprise rate limits