Continual-ARC scores lifelong skill acquisition on ARC-AGI streams
Harness persistence rarely beats a demonstration cache on memorization, and no harness method gains on unseen tasks, unlike weight updates.
Stories per day · 30 days
3 RSS
Relatedbenchmark3coding-agents3benchmarks2claude-code2github2self-evolution2agent-evaluation1agentmark1agents1anthropic1arc-agi1catastrophic-forgetting1codeql1continual-evolution1continual-learning1cursor1embodied-ai1evaluation1
41 stories, newest first · one card per story
Stories tagged #agent-harness
LongBanana benchmark tests long-context multi-reference image synthesis
LongBanana includes 90 tasks with 4,670 expert evaluation items for agent-driven image generation.
Continual harness evolution causes forgetting in coding agents
Meta-Harness drops 66% on SWE-Bench after evolution on FeatureBench with DeepSeek-V4-Flash-0731.
Auriga harness doubles GPT-6 Astra zero-shot robot success on RoboDojo
The frozen harness posts 57.7% success and stays first under scene randomization while learned policies fall by up to 33.9%.
Agentic Video Generation and Recursive Self-Improvement Studied
Agent harness coordinates video models for agentic generation, world modeling, and recursive self-improvement in multi-shot video production.
Trained agent policies carry harness preferences that break under interface changes
Reordering MiniWoB++ observation fields drops trained success from 85.0% to 19.2% without changing optimal decisions.
Paper frames work agents as model-plus-harness systems for verifiable tasks
It defines work persistence and openness, five work forms, and model, harness, and joint optimization.
TraceTrial-Bench measures AI agent harness reliability and failures
Across 3,420 runs, harness scores span 59.8% to 72.0%, and 31.1% succeed despite a failed criterion.
Essay says Claude Mods turn the agent harness toward the user
Anthropic ships six built-in Claude Code mods, and the author released an open-source ModSmith pack.
2
Study measures how harness and context change LLM vulnerability discovery
No single setup covered all 265 confirmed root causes, and CodeQL context helped only the smallest of four models.
HarnessAdapt-RL trains coding-agent policy and harness together
On a 9B model, pass@1 rises 9.8, 12.6, and 6.4 points over GRPO on three SWE-bench sets.
RiseHarness iteratively improves terminal-agent harnesses with weights fixed
It studies failed versus successful runs of the same task and adapts shared interaction mechanisms, not model weights.
Hacker News discusses Harness Engineering 101 for LLMs
The post has 57 points and 26 comments; the feed snippet gives no further technical detail.
Google Research posts Harness2, a recursive agent harness
The only available text is a Hacker News listing with 43 points and nine comments.
What Survives Self-Evolution: Provenance Marks in Copied Agent Harnesses
Provenance marks in leaked agent harnesses mostly fail when a self-evolving LLM rewrites the marked code.
Harness-delta method finds much agent-harness gain is overfitting
In 21 significant settings, overfitting and test-time scaling explain 54% and 14% of the gain on average.
Cortex: Measurement-Gated Evolution and Pruning of Coding-Agent Skills and Rules
Cortex keeps coding-agent skills and rules only when replayed tests show a net gain.
Learn2Play Bench: How Well Do LLM Agents Learn from Experience in Unfamiliar Environments?
Learn2Play Bench tests whether LLM agents learn novel game rules from interaction instead of pretrained knowledge.
Can a Cloud-Native Harness Make Agents Reliable Beyond the Desktop?
Kubernetes creators' Stacklok is building Mecatl, an open-source cloud-native harness to run coding agents beyond the desktop.
WebFovea: When the Model Is Right but the Click Is Wrong -- Reliable Round Trips for Vision-Based Web Agents on Live Websites
WebFovea placed second in WebRetriever 2026 after harness fixes lifted its score from 31.0 to 57.0.
Google researchers find a way to keep self-improving AI agents from memorizing their tests
Google's RRSI keeps self-improving agent harnesses from memorizing tests, adding up to 4.7 points on unseen benchmarks.
DeepSeek Harness v0.2 Brings Official Desktop Apps to Its Open-Source Agent Harness
DeepSeek released official macOS and Windows desktop apps for MIT-licensed DeepSeek Harness v0.2.
How Much of a Harness Does a Strong Agent Need for Autonomous ML Engineering?
Ablations show state-of-the-art MLE-agent harnesses offer no advantage over one minimal coding-agent session when the frontier LLM backbone and time budget are equal.
Nvidia's SoL-Pi system cuts coding agent token usage nearly in half by optimizing the harness
Nvidia's SoL-Pi research system cuts coding-agent token use nearly in half by optimizing the harness.