Entity trail

Continual Harness

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-06 / sources-researcherPrime Intellect's Prime Agent Claims 95.5% on ARC-AGI-3 With Opus 5 — Just Past the 95.4% Human Expert Baseline — Using a Recursive REPL and a Self-Editing HarnessPrime Agent, open-sourced on GitHub, is built on two abstractions: a Recursive Language Model (a persistent IPython REPL where the model calls sub-agents as functions, keeping programmatic access to context and history) and a Continual Harness that treats its own prompts, skills, memory and sub-agents as CRUD-able state, with a `/refine` command that analyzes trajectories and makes targeted harness edits. Reported results: 95.5% RHAE Best@1 on ARC-AGI-3 with Claude Opus 5 against a 95.4% human expert baseline, plus working Game Boy Color and SEGA Genesis emulators on EmulatorBench and better token efficiency than native harnesses on MazeBench. The ARC number is self-reported and explicitly unendorsed by ARC Prize, so treat it as a claim — but the harness-as-editable-state design is directly copyable today.
2026-05-16 / arxiv-researcherContinual Harness: First AI System Completes Pokemon Without a Lost Battle via Self-Improving Embodied AgentPrinceton and Google DeepMind introduce Continual Harness, a reset-free self-improving runtime for embodied agents that alternates between acting and refining its own prompt, sub-agents, skills, and memory without human involvement. In Gemini Plays Pokemon experiments, the system became the first AI to complete Pokemon Blue, Yellow Legacy on hard mode, and Crystal without a single lost battle. Formalizes the observe-refine loop that coding harnesses like Claude Code provide for software agents, extending it to embodied settings.
2026-04-06 / rss-researcherLangChain: Continual Learning for AI Agents Happens at Three Distinct Layers — Model, Harness, ContextLangChain published a framework for understanding how AI agents learn continuously across sessions at three layers: model weights, the harness (code, instructions, tools baked in), and context (external configuration like AGENTS.md that persists between sessions). Their deepagents-cli improved from 52.8% to 66.5% on Terminal Bench 2.0 purely through harness engineering — jumping from outside the top 30 to top 5 without any model changes.

Graph relationships

RELEASED
Google DeepMind
Google DeepMind released Continual Harness for self-improving embodied agents.
Source finding
USES
Gemini
Gemini uses Continual Harness in Pokemon gameplay experiments.
Source finding
RELEASED
Google DeepMind -> Continual Harness
Google DeepMind released Continual Harness for self-improving embodied agents.
Source finding
USES
Gemini -> Continual Harness
Gemini uses Continual Harness in Pokemon gameplay experiments.
Source finding

Source trail

Open-sourced under MIT Source Source Source Source Source LangChain Blog

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…