Awesome list for AI agent harness engineering: tools, patterns, evals, memory, MCP, permissions, observability, and orchestration.
Python agent-harnessagent-memoryagent-orchestrationai-agent-harnessai-agentsawesome-listcontext-engineeringharness-engineeringmcp
Add SmallCode: terminal-native coding agent harness optimized for 8B–35B local models
ai-boostcc08563f2017
diff --git a/README.md b/README.mdindex b188671..531db6e 100644--- a/README.md+++ b/README.md@@ -382,6 +382,7 @@ Real repositories worth studying — each with a note on *why* it's worth your t- [ClawGUI](https://github.com/ZJU-REAL/ClawGUI) — End-to-end harness for GUI agents that unifies online RL training, standardized benchmarking, and real-device deployment in one framework. The most complete open-source reference for building visual perception-action loops across desktop and mobile surfaces. - [desloppify](https://github.com/peteromallet/desloppify) — Agent harness that turns codebase quality improvement into a structured, score-driven workflow: mechanical detectors find dead code and complexity, LLM review assesses naming and abstractions, and a persistent `next` → fix → `resolve` loop keeps the agent on track across sessions. The anti-gaming score design — where wontfix items widen the gap between lenient and strict scores — is a rare example of harness-level incentive engineering that forces genuine improvement rather than metric hacking. - [DeepSeek-Reasonix](https://github.com/esengine/DeepSeek-Reasonix) — DeepSeek-native coding agent harness engineered around prefix-cache stability as a loop invariant: immutable-prefix / append-only-log / volatile-scratch partitioning achieves 99.82% cache-hit rates and ~5× cost reduction on long sessions. The most detailed public case study of designing an entire agent loop to preserve a provider-specific economic property rather than treating caching as an afterthought. +- [SmallCode](https://github.com/Doorman11991/smallcode) — Terminal-native coding agent built from the ground up for 8B–35B local models. Its harness innovations are all compensations for small-model limitations: 2-stage tool routing halves schema overhead, a forgiving multi-format parser recovers from malformed JSON/YAML/XML tool calls, patch-first editing avoids the truncation errors common when small models rewrite entire files, and a context budget engine with mid-turn eviction and semantic compression keeps long sessions inside narrow context windows. Demonstrates that harness design — not model scale — is the binding constraint for reliable agent behavior: it achieves 87% benchmark scores with a 4B-active model. ### Adjacent Collections