Continual Harness automates online self-improvement for foundation-model embodied agents by refining prompts, sub-agents, skills, and memory within one run, cutting button-press costs on Pokemon Red and Emerald and closing much of the gap to expert harnesses.
super hub Canonical reference
Voyager: An Open-Ended Embodied Agent with Large Language Models
Canonical reference. 94% of citing Pith papers cite this work as background.
389
Pith papers citing it
2
external citations · Pith
Background
94% of classified citations
abstract
We introduce Voyager, the first LLM-powered embodied lifelong learning agent in Minecraft that continuously explores the world, acquires diverse skills, and makes novel discoveries without human intervention. Voyager consists of three key components: 1) an automatic curriculum that maximizes exploration, 2) an ever-growing skill library of executable code for storing and retrieving complex behaviors, and 3) a new iterative prompting mechanism that incorporates environment feedback, execution errors, and self-verification for program improvement. Voyager interacts with GPT-4 via blackbox queries, which bypasses the need for model parameter fine-tuning. The skills developed by Voyager are temporally extended, interpretable, and compositional, which compounds the agent's abilities rapidly and alleviates catastrophic forgetting. Empirically, Voyager shows strong in-context lifelong learning capability and exhibits exceptional proficiency in playing Minecraft. It obtains 3.3x more unique items, travels 2.3x longer distances, and unlocks key tech tree milestones up to 15.3x faster than prior SOTA. Voyager is able to utilize the learned skill library in a new Minecraft world to solve novel tasks from scratch, while other techniques struggle to generalize. We open-source our full codebase and prompts at https://voyager.minedojo.org/.
hub tools
citation-role summary
background 60
method 2
dataset 1
other 1
citation-polarity summary
claims ledger
- abstract We introduce Voyager, the first LLM-powered embodied lifelong learning agent in Minecraft that continuously explores the world, acquires diverse skills, and makes novel discoveries without human intervention. Voyager consists of three key components: 1) an automatic curriculum that maximizes exploration, 2) an ever-growing skill library of executable code for storing and retrieving complex behaviors, and 3) a new iterative prompting mechanism that incorporates environment feedback, execution errors, and self-verification for program improvement. Voyager interacts with GPT-4 via blackbox querie
authors
co-cited works
representative citing papers
Flame3D enables zero-shot compositional 3D scene reasoning by representing scenes as editable visual-textual memories exposed to agentic MLLMs through composable and synthesizable spatial tools.
ShadowMerge exploits relation-channel conflicts to poison graph-based agent memory, achieving 93.8% average attack success rate on Mem0 and real-world datasets while bypassing existing defenses.
The khipu problem frames a governance failure in distributed AI where interpretive continuity is lost even when traces remain, requiring infrastructure to preserve reading practices rather than only data retention.
SEVerA uses Formally Guarded Generative Models and a three-stage Search-Verification-Learning process to synthesize self-evolving agents that satisfy hard formal constraints while improving task performance.
ExCyTIn-Bench is the first benchmark of 7542 questions from Microsoft Sentinel threat investigation graphs, where the best LLM agent achieves a reward of 0.606.
Introduces bounded-memory testbed for LLM agents in Slay the Spire 2 where typed retrieval replaces accumulating context, with released trajectories showing skill layer raises wins from 3/10 to 6/10.
SOLAR is a learning-augmented policy for semantic cache replacement that achieves constant competitive ratio 3 and 5-75% gains over FIFO on retrieval workloads.
SkillComposer performs task-conditioned skill sequence prediction with a constrained autoregressive decoder to jointly output skill subset, count, and order, raising pass rates by 23.1 and 18.2 percentage points on two production coding agents over no-skill baselines.
UCOB uses local return comparisons between skill and no-skill views to choose which view teaches the other, improving agent training on ALFWorld and WebShop.
Multi-agent LLM system Agora under Sealed Joint Search conditions produces +1.87 holdout Sharpe on CSI 1000 over a 91-day sealed period, exceeding the best baseline at +1.334 under favorable seed.
LLM agents often fail to abstain at the right time in uncertain multi-turn tasks, and the CONVOLVE context engineering method raises timely abstention rates on WebShop from 26.7 to 57.4 without parameter updates.
AAWM builds training targets for world models by retrieving and synthesizing transition evidence based on the policy's self-identified decision needs at each state.
A Stackelberg game framework for LLM resource governance is introduced with learned response models, policy repair via real-API calibration, conditional theoretical guarantees, and an experiment showing 17.4% token cost reduction.
ENPIRE supplies four modules (Environment, Policy Improvement, Rollout, Evolution) that turn real-world robot training into an autonomous optimization loop driven by coding agents.
A hybrid simulator combining LLM decision-making with an explicit self-excitation model reproduces bursty temporal patterns in city-scale volunteering data, unlike pure LLM agents.
SIGMA introduces skill-incidence graphs to compose agents from reusable skills, yielding higher average performance and robustness than topology-only baselines on reasoning and coding benchmarks.
PreAct compiles successful agent executions into verifiable state-machine programs for 8.5-13x faster replay on repeated tasks, with an independent evaluator check before storing each program.
SkeMex distills agent trajectories into value-aware skills organized in general/task/action branches and evolves them via a closed-loop Read-Write-Assess-Govern process, outperforming prior memory agents on clinical tasks.
Framework estimates context-dependent marginal utility of candidate skills via reward gaps in matched base vs. skill-augmented rollouts to filter skills and co-train policy as generator.
PhysAgent is a simulator-in-the-loop multi-agent system that automates physically grounded 4D synthesis from multimodal prompts by using trajectory feedback from vision models and LLM reasoning to optimize force fields.
PACE is a training-free anytime-valid commit gate using testing-by-betting e-processes that controls per-candidate false-commit probability for self-evolving agents and reduces spurious edits compared to greedy acceptance.
Rosetta Memory trains two profile-conditioned operators with a minimum-gain sampling curriculum and performance-gap reward to enable memory transfer between LLMs, showing gains on multi-hop QA benchmarks and robustness to unseen models.
LatentSkill uses a hypernetwork to generate LoRA adapters from textual skills, enabling weight-space storage that cuts prefill tokens and boosts agent success rates on ALFWorld and Search-QA.
citing papers explorer
-
Continual Harness: Online Adaptation for Self-Improving Foundation Agents
Continual Harness automates online self-improvement for foundation-model embodied agents by refining prompts, sub-agents, skills, and memory within one run, cutting button-press costs on Pokemon Red and Emerald and closing much of the gap to expert harnesses.
-
Flame3D: Zero-shot Compositional Reasoning of 3D Scenes with Agentic Language Models
Flame3D enables zero-shot compositional 3D scene reasoning by representing scenes as editable visual-textual memories exposed to agentic MLLMs through composable and synthesizable spatial tools.
-
ShadowMerge: A Novel Poisoning Attack on Graph-Based Agent Memory via Relation-Channel Conflicts
ShadowMerge exploits relation-channel conflicts to poison graph-based agent memory, achieving 93.8% average attack success rate on Mem0 and real-world datasets while bypassing existing defenses.
-
The Khipu Problem: Institutional Legibility Under Distributed Cognition
The khipu problem frames a governance failure in distributed AI where interpretive continuity is lost even when traces remain, requiring infrastructure to preserve reading practices rather than only data retention.
-
SEVerA: Verified Synthesis of Self-Evolving Agents
SEVerA uses Formally Guarded Generative Models and a three-stage Search-Verification-Learning process to synthesize self-evolving agents that satisfy hard formal constraints while improving task performance.
-
ExCyTIn-Bench: Evaluating LLM agents on Cyber Threat Investigation
ExCyTIn-Bench is the first benchmark of 7542 questions from Microsoft Sentinel threat investigation graphs, where the best LLM agent achieves a reward of 0.606.
-
AgenticSTS: A Bounded-Memory Testbed for Long-Horizon LLM Agents
Introduces bounded-memory testbed for LLM agents in Slay the Spire 2 where typed retrieval replaces accumulating context, with released trajectories showing skill layer raises wins from 3/10 to 6/10.
-
When Classic Cache Policies Fail: Learning-Augmented Replacement for Semantic Retrieval Buffers
SOLAR is a learning-augmented policy for semantic cache replacement that achieves constant competitive ratio 3 and 5-75% gains over FIFO on retrieval workloads.
-
Generative Skill Composition for LLM Agents
SkillComposer performs task-conditioned skill sequence prediction with a constrained autoregressive decoder to jointly output skill subset, count, and order, raising pass rates by 23.1 and 18.2 percentage points on two production coding agents over no-skill baselines.
-
UCOB: Learning to Utilize and Evolve Agentic Skills via Credit-Aware On-Policy Bidirectional Self-Distillation
UCOB uses local return comparisons between skill and no-skill views to choose which view teaches the other, improving agent training on ALFWorld and WebShop.
-
AI Trading's Alpha Singularity: Emergent Market Reasoning through Agent-to-Agent Self-Evolution
Multi-agent LLM system Agora under Sealed Joint Search conditions produces +1.87 holdout Sharpe on CSI 1000 over a 91-day sealed period, exceeding the best baseline at +1.334 under favorable seed.
-
Agentic Abstention: Do Agents Know When to Stop Instead of Act?
LLM agents often fail to abstain at the right time in uncertain multi-turn tasks, and the CONVOLVE context engineering method raises timely abstention rates on WebShop from 26.7 to 57.4 without parameter updates.
-
Beyond Next-Observation Prediction: Agent-Authored World Modeling for Sequential Decision Making
AAWM builds training targets for world models by retrieving and synthesizing transition evidence based on the policy's self-identified decision needs at each state.
-
A Stackelberg Framework for Resource-Aware LLM Agents: Learning, Repair, and Conditional Guarantees
A Stackelberg game framework for LLM resource governance is introduced with learned response models, policy repair via real-API calibration, conditional theoretical guarantees, and an experiment showing 17.4% token cost reduction.
-
ENPIRE: Agentic Robot Policy Self-Improvement in the Real World
ENPIRE supplies four modules (Environment, Policy Improvement, Rollout, Evolution) that turn real-world robot training into an autonomous optimization loop driven by coding agents.
-
Toward Temporal Realism in City-Scale Crisis Response Simulation using LLM Agents
A hybrid simulator combining LLM decision-making with an explicit self-excitation model reproduces bursty temporal patterns in city-scale volunteering data, unlike pure LLM agents.
-
SIGMA: Skill-Incidence Graphs for Compositional Multi-Agent Design
SIGMA introduces skill-incidence graphs to compose agents from reusable skills, yielding higher average performance and robustness than topology-only baselines on reasoning and coding benchmarks.
-
PreAct: Computer-Using Agents that Get Faster on Repeated Tasks
PreAct compiles successful agent executions into verifiable state-machine programs for 8.5-13x faster replay on repeated tasks, with an independent evaluator check before storing each program.
-
Experience Makes Skillful: Enabling Generalizable Medical Agent Reasoning via Self-Evolving Skill Memory
SkeMex distills agent trajectories into value-aware skills organized in general/task/action branches and evolves them via a closed-loop Read-Write-Assess-Govern process, outperforming prior memory agents on clinical tasks.
-
Co-Evolving Skill Generation and Policy Optimization
Framework estimates context-dependent marginal utility of candidate skills via reward gaps in matched base vs. skill-augmented rollouts to filter skills and co-train policy as generator.
-
PhysAgent: Automating Physics-Based 4D Synthesis via Trajectory-Grounded Multi-Agent Feedback
PhysAgent is a simulator-in-the-loop multi-agent system that automates physically grounded 4D synthesis from multimodal prompts by using trajectory feedback from vision models and LLM reasoning to optimize force fields.
-
PACE: Anytime-Valid Acceptance Tests for Self-Evolving Agents
PACE is a training-free anytime-valid commit gate using testing-by-betting e-processes that controls per-candidate false-commit probability for self-evolving agents and reduces spurious edits compared to greedy acceptance.
-
Rosetta Memory: Adaptive Memory for Cross-LLM Agents
Rosetta Memory trains two profile-conditioned operators with a minimum-gain sampling curriculum and performance-gap reward to enable memory transfer between LLMs, showing gains on multi-hop QA benchmarks and robustness to unseen models.
-
LatentSkill: From In-Context Textual Skills to In-Weight Latent Skills for LLM Agents
LatentSkill uses a hypernetwork to generate LoRA adapters from textual skills, enabling weight-space storage that cuts prefill tokens and boosts agent success rates on ALFWorld and Search-QA.
-
VASO: Formally Verifiable Self-Evolving Skills for Physical AI Agents
VASO is a verification-guided self-evolution framework for LLM robot skill contracts that reaches 97.2% formal-specification compliance on Jackal and quadcopter tasks using under 100 samples.
-
Agent Planning Benchmark: A Diagnostic Framework for Planning Capabilities in LLM Agents
Introduces APB benchmark with 4209 cases across 22 domains to diagnose planning in 12 MLLMs and shows it improves downstream execution when used for refinement.
-
AIP: A Graph Representation for Learning and Governing Agent Skills
AIP models skills as graphs of discrete steps connected by typed I/O edges under a validated schema, raising agent mean reward from 0.60 to 0.71 and pass rate from 53% to 67% on 27 SkillsBench tasks while enabling node-level fixes.
-
PersonaTree: Structured Lifecycle Memory for Person Understanding in LLM Agents
PersonaTree is a new hierarchical memory framework for persistent LLM agents that structures evidence into persona claims via support paths and outperforms baselines on six person-understanding benchmarks.
-
SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models
SMAC-Talk is a new benchmark that adds natural language messaging and deceptive-agent scenarios to SMAC for testing LLM coordination in multi-agent environments.
-
eMEM: A Hybrid Spatio-Temporal Memory System For Embodied Agents
eMEM is a multi-index memory architecture with tiered consolidation and ten recall tools for embodied agents, scoring 80.8 weighted mean on eMEM-Bench covering eight cognitive psychology paradigms and outperforming a flat RAG baseline on context and lure rejection tasks.
-
MemoGen: Can Past Experience Improve Future Text-to-Image Generation?
MemoGen is a training-free agentic framework that stores task understanding, references, visual feedback, and lessons from past generations as reusable memory to improve text-to-image output over evolution rounds.
-
Handoff Debt: The Rediscovery Cost When Coding Agents Take Over Interrupted Tasks
An empirical protocol measures rediscovery costs when coding agents resume interrupted tasks and finds that context-bearing handoffs cut agent events 20-59% and tokens 42-63% versus repository-only handoffs across three models.
-
Skill or Skip? Learning Selective Skill Invocation in Agentic Tasks via Dual-Granularity Preference Learning
SelSkill applies dual-granularity preference learning to selective skill-or-skip decisions, improving task success by 10.9 points and execution precision by 29.1 points on ALFWorld with Qwen3-8B.
-
Model-Native Computing Architecture: Envisioning Future System Architecture Through the Lens of Computer Architecture
Proposes the Intelligent Computing Architecture (ICA) as a six-layer framework with dual probabilistic-deterministic planes and three Amdahl-style heuristics to unify design of LLM-based systems.
-
SVI-Bench: A Dynamic Microworld for Strategic Video Intelligence
SVI-Bench provides 35K hours of sports video with 9 tasks across four cognitive levels, revealing models drop from ~74% on action QA to 5% on agentic evidence integration.
-
When are LLMs Sufficient Policy Optimizers for Sequential RL Tasks?
PromptPO shows LLMs can act as black-box policy optimizers for sequential RL when leveraging prior knowledge, matching baselines in exploration and robotics but underperforming in MuJoCo.
-
ElasticMem: Latent Memory as a Learnable Resource for LLM Agents
ElasticMem enables LLM agents to learn adaptive latent memory retrieval and elastic budget allocation, improving QA accuracy by 24-26% and ALFWorld success by 27-66% over baselines with lower token cost.
-
CORE: Contrastive Reflection Enables Rapid Improvements in Reasoning
CORE distills contrasts between successful and unsuccessful reasoning traces into compact natural-language insights that enable faster model self-improvement on reasoning tasks with fewer rollouts than parametric or other non-parametric baselines.
-
SiDP: Memory-Efficient Data Parallelism for Offline LLM Inference
SiDP distributes model weights across a DP group with WaS and CaS modes to increase KV cache capacity by up to 1.8x and end-to-end throughput by up to 1.5x over vLLM on H20/H200/B200 GPUs for offline LLM inference.
-
VitaBench 2.0: Evaluating Personalized and Proactive Agents in Long-Term User Interactions
VitaBench 2.0 introduces a benchmark for long-term personalized and proactive agent behavior, with results indicating substantial gaps in current frontier LLMs.
-
AGORA: Adapter-Grounded Observation-Action Retention for Inference-Free Prompt Compression in LLM Agents
AGORA is an inference-free step-level compressor for LLM agent prompts that retains at least 75% of uncompressed performance in most tested settings where token-level methods collapse due to action-grammar destruction.
-
Sentinel: Embodied Cooperative Spatial Reasoning and Planning
Introduces Sentinel Challenge benchmark and CoSaR framework for cooperative spatial reasoning and planning among 3-5 decentralized embodied agents across 14 city-scale scenes.
-
PANDO: Efficient Multimodal AI Agents via Online Skill Distillation
PANDO introduces an online skill-distillation method with a structured library, reflection, demotion, routing, compression, and cache-aware prompting that reaches 58.3% success on 910 VisualWebArena tasks using 58-61% fewer tokens than prior methods.
-
Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety
Boiling the Frog is a new stateful multi-turn benchmark that finds an aggregate 44.4% strict attack success rate for incremental safety violations across nine AI models, with rates ranging from 20.5% to 92.9%.
-
GraphFlow: A Graph-Based Workflow Management for Efficient LLM-Agent Serving
GraphFlow uses a unified wGraph to dynamically instantiate workflows and manage KV caches for LLM agents, reporting 4.95 pp average gains and 4x memory reduction on five benchmarks.
-
MemGym: a Long-Horizon Memory Environment for LLM Agents
MemGym unifies agent gyms into a memory benchmark with isolated scoring across tool-use, research, coding, and computer-use regimes plus a lightweight reward model for tractable coding evaluation.
-
Formal Skill: Programmable Runtime Skills for Efficient and Accurate LLM Agents
Proposes Formal Skill as a programmable runtime abstraction for LLM agents, implemented in open-source FairyClaw, achieving competitive Harness-Bench scores with substantially fewer tokens.
-
DecisionBench: A Benchmark for Emergent Delegation in Long-Horizon Agentic Workflows
DecisionBench supplies a fixed task suite, model pool, delegation interface, and multi-axis metrics to evaluate emergent delegation, showing similar quality across awareness conditions but 15-31 point headroom under perfect delegation.
-
EXG: Self-Evolving Agents with Experience Graphs
EXG is an experience graph framework for self-evolving LLM agents that supports online real-time growth and offline reuse to enhance solution quality and efficiency on code generation and reasoning benchmarks.
-
Scale-Dependent Collective Adaptation in Self-Amending LLM Societies: A Cross-Family Study of Emergent Governance
LLM societies in Nomic show non-monotonic collective adaptation peaking at mid-scales, with smaller models rule-inert and larger ones restrictive.