09:58
2026-10-13
stackness.dev
ai-agents
When more context makes your coding agent worse: subagent, /compact, /clear or a headless loop
Tool output accounted for 56% of context tokens and tool calls another 28% in a 22 September 2026 CMU study of coding agents on Terminal-Bench 2.0, according to the CliffCompaction paper from Tim Dett…