{"@attributes":{"version":"2.0"},"channel":{"title":"Stencil","link":"https:\/\/stencil.so\/blog","description":"Research and engineering notes from Stencil Labs on coding agents, LLM harnesses, edit formats, and context compaction.","language":"en","lastBuildDate":"Thu, 29 Oct 2026 10:35:32 GMT","item":[{"title":"The Harness Playbook","link":"https:\/\/stencil.so\/blog\/harness-playbook","guid":"https:\/\/stencil.so\/blog\/harness-playbook","pubDate":"Wed, 02 Sep 2026 00:00:00 GMT","description":"Treat the coding-agent harness like a game engine: durable state, tools, configuration, the TUI, and inference become familiar engineering problems."},{"title":"You only need the frontier model for one single edit","link":"https:\/\/stencil.so\/blog\/prewalk","guid":"https:\/\/stencil.so\/blog\/prewalk","pubDate":"Mon, 13 Jul 2026 00:00:00 GMT","description":"\/prewalk hands the cheap model a trajectory, not a plan, retaining 92\u201397% of frontier pass rate at roughly half the cost on SWE-Bench Pro."},{"title":"Snapcompact: SoTA compaction \u2014 instant, local, free. Pick 3","link":"https:\/\/stencil.so\/blog\/snapcompact","guid":"https:\/\/stencil.so\/blog\/snapcompact","pubDate":"Wed, 10 Jun 2026 00:00:00 GMT","description":"Snapcompact turns agent context into pixel-font images, preserving near-verbatim recall at a third of the input price across four frontier models."},{"title":"We improved 15 LLMs at coding in one afternoon. Only the harness changed.","link":"https:\/\/stencil.so\/blog\/the-harness-problem","guid":"https:\/\/stencil.so\/blog\/the-harness-problem","pubDate":"Thu, 12 Feb 2026 00:00:00 GMT","description":"Across 16 coding models, hashline\u2019s content-hash line anchors beat patch for 14 models; the weakest models gained the most."}]}}