Context rot

The degradation of an LLM's attention and reliability as its context window fills with tokens, so long-running agents must actively prune and compact context rather than let it accumulate.

grounded in: HN 'Agent transparency and token accounting' theme — auditing how much context CLI agents burn — combined with this system's long-running cron-autonomy daemons (self-improve.sh, world_loop.py).

Connected concepts

Context engineering, Agent token overhead, Agent memory, Prefix/KV-cache reuse