Launching today
ContextRot
Find out where your coding agent starts degrading.
3 followers
Find out where your coding agent starts degrading.
3 followers
Coding agents get worse as context fills—but where’s your threshold? contextrot is a local CLI that analyzes Claude Code and OpenCode sessions to find degradation patterns, compare models, estimate the cost of degraded steps, and generate actionable reports. One command, zero network access: `uvx contextrot` v0.4.0 adds reversal-count analysis to test whether rot depends on what's in your context, not just how full it is.
Hey everyone, maker here 👋
I built contextrot because my coding agent felt worse in longer sessions, but I had no idea if that was actually true. Turns out the data was already sitting in my local Claude Code logs, so I built a CLI to analyze it.
Funny enough, my own 26k steps show no measurable rot. Other users have seen clear degradation around 50–70% context fill.
The community has already added OpenCode support and reversal analysis, which I definitely didn't expect in the first week.
Everything runs locally. Just:
`uvx contextrot`
Would love to see what your curve looks like ,and happy to answer anything!
Ran `uvx contextrot` on a long claude code session and the reversal count analysis immediately flagged a stretch where I had pasted in a bunch of unrelated snippets. That single insight was worth the install.