Ouroboros: A Self-Developing Frontier Coding Agent with Reviewed Core Evolution

Hugging Face Daily Papers Papers
self-developing-agent coding-agent core-evolution benchmark frontier-ai autonomous-agents

Summary

Ouroboros is a self-developing frontier coding agent whose tools, prompts, and core implementation improve through reviewed commits, achieving state-of-the-art results on Terminal-Bench, OSWorld, and CL-Bench, with a long-running live deployment called Hope.

We present Ouroboros, a self-developing agent harness whose tools, prompts, context assembly, and core implementation improve through reviewed commits that become the runtime for later work. Core evolution proceeds in two modes. In recursive free evolution, improvement is itself a task, and completing one evolution cycle can schedule the next. In experience-driven core evolution, ordinary work and social interaction expose bugs, rough edges, and inefficient context construction that lead to reviewed structural changes. On Terminal-Bench 2.1, an Opus 5 run scores 86.74%, the best result reported on the benchmark. On OSWorld-Verified, an Opus 5 run reaches 90.69%, exceeding the best previously reported score. A five-rollout CL-Bench campaign achieves a normalized reward of 0.2301, setting a new state of the art. Hope is the longest-running publicly documented Ouroboros deployment. It is a 161-day living agent experiment in free evolution under governed human communication across seven surfaces. Human interaction surfaces faults and generates proposals, but the agent decides which changes to pursue. Because a self-developing agent may rewrite its own code and select new model APIs, operational safety becomes a primary design problem: guardrails must remain authoritative under evolutionary and public social pressure. Benchmark campaigns use frozen system snapshots, while Hope continues live evolution on a separate lineage.
Original Article
View Cached Full Text

Cached at: 08/11/26, 06:19 AM

Paper page - Ouroboros: A Self-Developing Frontier Coding Agent with Reviewed Core Evolution

Source: https://huggingface.co/papers/2608.08311

Abstract

WepresentOuroboros,aself-developingagentharnesswhosetools,prompts,contextassembly,andcoreimplementationimprovethroughreviewedcommitsthatbecometheruntimeforlaterwork.Coreevolutionproceedsintwomodes.Inrecursivefreeevolution,improvementisitselfatask,andcompletingoneevolutioncyclecanschedulethenext.Inexperience-drivencoreevolution,ordinaryworkandsocialinteractionexposebugs,roughedges,andinefficientcontextconstructionthatleadtoreviewedstructuralchanges.OnTerminal-Bench2.1,anOpus5runscores86.74%,thebestresultreportedonthebenchmark.OnOSWorld-Verified,anOpus5runreaches90.69%,exceedingthebestpreviouslyreportedscore.Afive-rolloutCL-Benchcampaignachievesanormalizedrewardof0.2301,settinganewstateoftheart.Hopeisthelongest-runningpubliclydocumentedOuroborosdeployment.Itisa161-daylivingagentexperimentinfreeevolutionundergovernedhumancommunicationacrosssevensurfaces.Humaninteractionsurfacesfaultsandgeneratesproposals,buttheagentdecideswhichchangestopursue.Becauseaself-developingagentmayrewriteitsowncodeandselectnewmodelAPIs,operationalsafetybecomesaprimarydesignproblem:guardrailsmustremainauthoritativeunderevolutionaryandpublicsocialpressure.Benchmarkcampaignsusefrozensystemsnapshots,whileHopecontinuesliveevolutiononaseparatelineage.

View arXiv pageView PDFProject pageGitHub1.06kAdd to collection

Get this paper in your agent:

hf papers read 2608\.08311

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2608.08311 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2608.08311 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2608.08311 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

State of CLI Coding Agents, Mid-2026 (37 minute read)

TLDR AI

A detailed comparison of CLI coding agents including Claude Code, Codex CLI, Omp, and OpenCode, noting that the top three produce similar quality results while OpenCode lags but works with various models.

@omarsar0: https://x.com/omarsar0/status/2068008743153832264

X AI KOLs Following

The article explains the shift from manually prompting coding agents to designing automated loops that prompt them, detailing what these loops are, their historical evolution, and the components needed to build them in production.