Grawkus NHE is the Grawkus-branded 3.0.0 base for observability-driven coding-agent harness evolution.
This repo is based on the Agentic Harness Engineering project from china-qijizhifeng/agentic-harness-engineering, with the Grawkus identity, package metadata, and command surface layered on top.
- A Python/uv harness-evolution workspace, not the old npm TypeScript CLI.
- A NexAU-based loop for evaluating, analyzing, and improving coding-agent harness components.
- A reproducible base for evolving prompts, tools, middleware, skills, sub-agents, and long-term memory.
- A Grawkus 3.0.0 reset point for future benchmark and harness work.
Requirements:
- Python 3.13 or newer
uvtmuxfor the upstream shell launch scripts- E2B credentials for sandboxed rollouts
uv syncCreate local environment config:
cp .env.example .envAt minimum, configure:
LLM_API_KEYLLM_BASE_URLE2B_API_KEYSERPER_API_KEY
After installation, the package exposes:
grawkus
grawkus-nhe
agentic-harness-engineeringEach entry point runs the same evolution orchestrator in evolve.py.
See BASE_MODEL.md for the exact upstream commit and local import notes.
This fork preserves the Agentic Harness Engineering architecture and upstream attribution. The original project is available at:
https://github.com/china-qijizhifeng/agentic-harness-engineering