Claude Mods: Flipping the Harness Inside Out

Claude Mods are an early implementation of the Attention Interface that lets users customize Claude Code—the agent harness—using TypeScript or JavaScript plugins. These mods act as middleware, intercepting events (prompts, tool calls, renders) to alter how the interface behaves and appears at runtime. By reshaping the human-facing portion of the harness, they enable a platform‑like environment where companies can build on top of the core model engine while controlling the agent’s interaction surface.

Claude Code’s Next Era — Thariq Shihipar, Anthropic
Impressive paper showing how much the harness changes a coding agent's results. Harnesses do play a huge role in what you are getting out of…
If you build agent harnesses, this is important. Should you avoid subagents, or can they be useful? My thoughts as a harness builder: I reme…
No Claude sub, and I'm still getting the 'Claude Code experience' by routing OpenAI, Copilot, GLM, and Ollama through one harness
Strong Opinions, Loosely Held on Agent + Harness Engineering: 1. You can outperform any default harness+model (including codex & claude code…

Source: Latent Space | 2026-10-21

Related