Long AI coding-agent sessions build up clutter in their context window: a file read twice, a tool result that's stale after a later edit, a system reminder repeated over and over, a huge log that's skimmed once and never looked at again. contextrot reads a session transcript and measures how much of it is still useful ("live") versus dead weight ("rot"). It reports token totals by rot category and estimates how many turns are left before the conversation runs out of context window and gets compacted.
This page runs entirely in your browser. Nothing you paste here is uploaded anywhere.
| Category | Count | Tokens |
|---|
| # | Category | Tokens | Summary |
|---|