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 FrontierAgent to Demo Harnesses
ai-boost84c9f443df27
diff --git a/README.md b/README.mdindex e976f53..c97f7b0 100644--- a/README.md+++ b/README.md@@ -495,6 +495,7 @@ Real repositories worth studying — each with a note on *why* it's worth your t- [Pi](https://github.com/earendil-works/pi) — Minimal terminal coding harness built around "lazy skills": each capability keeps only a one-line description in active context, loading full instructions and tool schemas only when invoked. Keeps the system prompt under 1,000 tokens versus 7,000–10,000 for typical agents, making it a concrete reference for context-minimal harness design. - [grok-build](https://github.com/xai-org/grok-build) — SpaceXAI's open-source terminal coding agent harness: a Rust-based fullscreen TUI with an extensible tool runtime, MCP/skills/hooks support, and headless/embedded modes via ACP. A useful first-party counterpoint to Claude Code and Codex CLI for studying how a new model provider structures the loop, checkpoints workspace state, and exposes the same harness through interactive and scripted surfaces. - [CodeWhale](https://github.com/Hmbown/CodeWhale) — Open-source, community-driven terminal coding agent harness written in Rust with a model-agnostic runtime, OS-level sandbox (Seatbelt/Landlock/seccomp/bwrap), resumable fleets via an append-only ledger, and a `/model` command that lets you switch providers mid-task. A strong example of treating the harness — not the model — as the product surface, with clear separation between interactive TUI, headless `exec`, and embedded web modes. +- [FrontierAgent](https://github.com/ApodexAI/FrontierAgent) — Apodex's open-source agent runtime (August 2026) shipping two native workflows — ReAct and Agent Team (a coordinator maintains a live task board, delegates bounded work to parallel sub-agents, and synthesizes their reports). The design decisions are the value: a task-scoped filesystem contract (`/inputs` read-only, `/workspace` mutable, `/outputs` as persistent deliverables) with fail-closed authorization, asynchronous intervention injected at safe turn boundaries without discarding the active run, and checkpoint/trace/revert built into every session. The same workflow engine powers the bundled benchmark runner — a concrete demonstration that the agent harness and its eval harness should be one engine, not two systems that drift apart. - [DeepSeek Harness (dsh)](https://github.com/deepseek-ai/deepseek-harness) — DeepSeek's official open-source agent harness (August 2026) built on an "everything is a plugin" architecture powered by Cordis: models, tools, skills, UI surfaces, and even the desktop shell are plugins composed through dependency injection rather than a monolith with extension points, with an accompanying paper on spatiotemporal composability. The clearest first-party reference for treating harness extensibility as the primary design axis instead of an afterthought — and at 200K+ stars in its first month, the strongest signal of where the community's center of gravity is moving. ### Adjacent Collections