Prime Intellect Harness

I want to evaluate and, if it fits, integrate a technique from recent AI research into my project.

## Technique: Prime Intellect Harness
Prime Intellect open-source agent harness boosts ARC-AGI-3 from 30% to 95.5%

Drop-in applicable open-source agent harness for context reuse across runs (Prime Intellect, AlphaSignal signal · 2026-08-26). Prime Intellect open-sourced a harness that lets models reuse context across agent runs — context accumulated in one run persists into the next rather than being reset. The result on ARC-AGI-3: score jumps from 30% (stateless baseline) to 95.5% (persistent-context harness) — a 65+ point improvement. The harness separates orchestration (what to try next) from execution (how to try it), preserving the full reasoning trace across session boundaries. The persistent-context pattern is directly applicable to any production agent system, long-horizon autonomous agent, or iterative RAG pipeline where reasoning context should compound across calls rather than reset.

What the reference demo shows: Gradio demo with three sub-tabs — (1) Context Reuse Demo: stateless agent vs. persistent-context agent side-by-side on an abstract reasoning task (ARC-like grid), showing how accumulated context leads to correct solution vs. repeated wrong attempts; (2) Score Progression: interactive chart of ARC-AGI-3 improvement from 30% to 95.5% as harness features are added (context reuse, orchestrator/executor split, structured disagreement); (3) Architecture Guide: drop-in implementation pattern, when context reuse matters most, production deployment checklist.

## Reference implementation (Vector Ridge Labs, POC-025)
- Live demo: https://osoman2-gradio-pocs.hf.space/?poc=POC-025
- Source code: https://huggingface.co/spaces/osoman2/Gradio_pocs/tree/main/prime_intellect_harness
Original source: AlphaSignal signal · 2026-08-26

## My project
<Describe your stack, the problem you want to solve and where this technique would sit. Example: "Python FastAPI RAG service over 50k PDFs, Postgres + pgvector, OpenAI embeddings, p95 latency budget 800 ms.">

## What I need from you
1. Read the reference source code and explain in at most 5 bullets how the technique works and what it would change compared to my current approach.
2. Decide whether it fits my case. If it does not, say why and stop there.
3. Propose the smallest viable integration: files or modules to touch, new dependencies, configuration.
4. Implement it behind a flag or as a separate function so it can be compared side by side with what I have today.
5. Add an evaluation script that measures task success rate, number of steps and token cost per task, before and after, on my own data.
6. List the risks (latency, cost, failure modes) and what to monitor once it is in production.

Constraints: keep my existing stack and conventions, avoid heavy dependencies unless the gain is clear, and reference the file in the demo source whenever you reuse its logic.