Research
Blog
The Harness PlaybookTreat the coding-agent harness like a game engine: durable state, tools, configuration, the TUI, and inference become familiar engineering problems.You only need the frontier model for one single edit/prewalk hands the cheap model a trajectory, not a plan, retaining 92–97% of frontier pass rate at roughly half the cost on SWE-Bench Pro.Snapcompact: SoTA compaction — instant, local, free. Pick 3Snapcompact turns agent context into pixel-font images, preserving near-verbatim recall at a third of the input price across four frontier models.We improved 15 LLMs at coding in one afternoon. Only the harness changed.Across 16 coding models, hashline’s content-hash line anchors beat patch for 14 models; the weakest models gained the most.