Type: url
Original: https://www.humanlayer.dev/blog/context-efficient-backpressure
Status: partial
Summary
A short HumanLayer blog post from December 9, 2025 recommending that test, build and lint output be swallowed and replaced with a single check mark when a stage passes, with the full output shown only on failure, so coding agents do not waste context on passing results. It gives a run_silent shell wrapper, suggests fail-fast flags, output filtering and framework-specific parsing, and criticizes recent models’ own habits of discarding output or piping it to head/tail, which it says end up costing more tokens and human time.