https://stencil.so/blog Research and engineering notes from Stencil Labs on coding agents, LLM harnesses, edit formats, and context compaction. en Sun, 04 Oct 2026 07:08:22 GMT https://stencil.so/blog/harness-playbook https://stencil.so/blog/harness-playbook Wed, 02 Sep 2026 00:00:00 GMT Treat the coding-agent harness like a game engine: durable state, tools, configuration, the TUI, and inference become familiar engineering problems. https://stencil.so/blog/prewalk https://stencil.so/blog/prewalk Mon, 13 Jul 2026 00:00:00 GMT /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. https://stencil.so/blog/snapcompact https://stencil.so/blog/snapcompact Wed, 10 Jun 2026 00:00:00 GMT Snapcompact turns agent context into pixel-font images, preserving near-verbatim recall at a third of the input price across four frontier models. https://stencil.so/blog/the-harness-problem https://stencil.so/blog/the-harness-problem Thu, 12 Feb 2026 00:00:00 GMT Across 16 coding models, hashline’s content-hash line anchors beat patch for 14 models; the weakest models gained the most.