Consistent Plan-Act for Long-Horizon Agentic Tasks

Heng-Zhuang Li; Yi-Kai Zhang; Yu Wang; Yueqing Sun; Jiayuan Zhang; Qi Gu; Han-Jia Ye

cs.AISep 30, 2026

Consistent Plan-Act for Long-Horizon Agentic Tasks

Authors: Heng-Zhuang Li, Yi-Kai Zhang, Yu Wang, Yueqing Sun, Jiayuan Zhang, Qi Gu, Han-Jia Ye

Organizations: School of Artificial Intelligence, Nanjing University · National Key Laboratory for Novel Software Technology, Nanjing University · LongCat Team, Meituan · University of Science and Technology of China

Abstract

Long-horizon agentic tasks demand strong reasoning and efficient execution across successive interactions with dynamic environments. A common approach decouples high-level planning from low-level execution through separate planner and actor roles. To investigate coordination failures in these tasks, we prompt both agents for structured state assertions and compare their reports programmatically to detect explicit contradictions. Our analyses reveal systematic disagreement about the same task-relevant state facts, a phenomenon we term planner-actor state mismatch. We further find that providing agents with task-relevant state information reduces mismatch and improves coordination and task performance. Based on the systematic analysis of the state mismatch, we propose Consistent Plan-Act (ConPAct), which feeds detected contradictions back to both agents to form consistent state interpretations and fine-tunes them on curated consistent interactions for better coordination. ConPAct improves performance across various environments and model configurations, e.g., increasing MiniGrid success rate from 38.6% to 54.4% with GPT-5.6-sol/terra as planner and actor respectively, demonstrating that state consistency can guide both inference-time correction and coordination training.

Figures & tables

Panel of Figure 1 — Figure 1: Planner-actor state mismatch . (a) Misreading the upper box as one column farther right, the actor pushes Right twice, causing deadlock; the correct sequence is Right $\rightarrow$ Up , with consistent interpretation as the planner. (b) Mismatch occurs in 85.0% of failed trajectories, far exceeding non-adherence (31.1%) and low planning/execution capability (8.4%).

Panel of Figure 2 — Figure 2: Prevalence of planner-actor state mismatch. (a) Mismatch occurs across various model configurations, with even homogeneous configurations have notable state mismatch; (b) State mismatch coexists with high sub-goal adherence, with marker size indicating trajectory frequency. Model sources are provided in Appendix C.2 .

Panel of Figure 3 — Figure 3: Planner-actor state mismatch is systematic and prevalent in failed trajectories. (a) Mismatch persists under both visual and textual observations; (b) mean cross-role disagreement exceeds within-role sampling variability; (c) mismatch-only failures dominate across task configurations.

Panel of Figure 4 — Figure 4: State information and consistent plan-act coordination. (a) Task-relevant state facts improve the actor’s decisions. (b) Injecting reliable state assertions reduces mismatch and improves task success. (c) ConPAct feeds detected contradictions back for coordination reconciliation.

Panel of Figure 5 — Figure 5: OSWorld performance and Sokoban inference-time comparisons. (a) OSWorld Score and SR relative to the stronger of ReAct and Plan-Act within each model, measured in percentage points; point labels give absolute percentages and purple annotations give ConPAct-I gains. (b) Task success (bars, left axis) and disagreement repair (lines, right axis) under different correction settings. (c) Task success with assertion prompting, resampling, Self-Refine, and ConPAct-I.

Panel of Figure 6 — Figure 6: Supervised-training ablations and state analysis on Sokoban. (a) Task success and mismatch under ConPAct-R data ablations; Full includes consistency filtering, reconciliation supervision, and reconciliation replay. (b) Initial mismatch and assertion correctness; annotations show changes in percentage points from the baseline. (c) Discussion and repair rates for ConPAct-R and its data ablations. The components together contribute to the effectiveness of ConPAct.

Appendix figures & tables20 assets

Supplementary material from the paper’s appendix.

Appendix

Game	Target for target_visible	Threat for hazard_nearby
bigfish	At least one visibly smaller fish	A larger fish
chaser	At least one uncollected green orb	A non-vulnerable enemy
coinrun	The goal coin	A saw, enemy, or deadly gap
jumper	The carrot	Spikes
maze	The cheese	Not included in the schema
miner	At least one diamond	A boulder or diamond that could fall onto the player

Appendix

Table 3: Game-specific referents for Procgen assertions. Maze has no hazard_nearby field.

Environment	Source instances	Evaluation instances	Evaluation rollouts	$A$	$C$
Sokoban	100	200	1,600	60	60
Crafter	100	200	1,600	160	160
Procgen	100	140	1,120	160	160
MiniGrid / BabyAI	108	150	1,200	60–200	60–200

Appendix

Table 4: Environment splits and evaluation budgets.

A

limits executed actions and

C

limits total logical model calls per rollout. Each instance has eight evaluation rollouts. MiniGrid budgets are specified per task in Table 6 .

Game	Source levels	Evaluation levels	$R_{\min}$	$R_{\max}$
bigfish	14	20	1	40
chaser	14	20	0.5	13
coinrun	15	20	5	10
jumper	14	20	3	10
maze	14	20	5	10
miner	15	20	1.5	13

Appendix

Table 5: Procgen task allocation and easy-mode normalization bounds from Cobbe et al. [2020] . Each game has 20 evaluation levels with IDs 100–119. Source levels are selected from IDs 0–49.

Task ID	Instances	Seed offsets	$A=C$
BabyAI-BlockedUnlockPickup-v0	5	0, 1, 2, 3, 4	136
BabyAI-GoTo-v0	5	0, 1, 2, 3, 4	200
BabyAI-GoToDoor-v0	5	0, 1, 2, 3, 4	84
BabyAI-GoToLocal-v0	5	0, 1, 2, 3, 4	60
BabyAI-GoToObj-v0	5	0, 1, 2, 3, 4	60
BabyAI-GoToRedBall-v0	5	0, 1, 2, 3, 4	60

Appendix

Table 6: MiniGrid and BabyAI evaluation tasks. Seed offsets are added to 1000; each resulting instance has eight rollouts. The last column gives the common action and model-call limit for each rollout.

Setting	Value
Trainable parameters	Language-model backbone; vision encoder/projector frozen
Reference schedule	3 epochs of ReAct SFT; token-matched across methods
Training seed	42
Observation window	3 frames, including the current observation
Optimizer	AdamW (fused)
Learning rate	$1\times 10^{-5}$

Appendix

Table 7: Supervised fine-tuning settings shared by the training methods. Token budgets are matched to the three-epoch ReAct SFT reference for each environment and model size.

Method	Supervised targets
ReAct / Plan-Act / ECoT	Complete valid responses from retained demonstrations or progress prefixes.
PAA	The planner’s annotated plan and the actor’s action-only responses.
HSL	Expert responses and goal-relevant actions from hindsight examples; irrelevant actions are excluded.
WebSTAR	Original responses for actions with judge scores greater than 5; other responses are excluded.
ConPAct-S	Valid responses from consistent segments and successful corrections; conflicting replies remain context only.
ConPAct-R	Validated responses under the same consistency rule; hypothetical errors and recorded nonoptimal actions are excluded from targets.

Appendix

Table 8: Response selection for supervised training. Loss is applied only to the retained target response.

Method	Planner input	Actor input
ReAct / HSL / WebSTAR	No separate planner	Up to 3 recent frames
Plan-Act / Plan-Act SFT	Current frame plus up to 2 sub-goal-start frames	Up to 3 recent frames
PAA	Initial frame; called once	Up to 3 recent frames
ECoT	Current frame plus up to 2 sub-goal-start frames	Up to 3 recent frames
HiPlan	Initial frame for the global plan; up to 3 recent frames for hints	Up to 3 recent frames
TAPE	Up to 3 recent frames for visual calls	Up to 3 recent frames

Appendix

Table 9: The common three-frame observation limit. A planner called only at initialization receives the single available frame. Textual task instructions and active plans are retained separately.

Figure 7: Paired Sokoban observations for the initial state of level Stage_1_5_2387 . The archived screenshot and the text generated from the recorded board show the same two boxes, two targets, player position, and wall layout. Neither box is on a target.

Model

Method

Chrome

GIMP

Calc

Impress

Writer

Multi- apps

Thunder- bird

VLC

VS Code

Overall

Opus-5

ReAct

71.0/69.3

87.2/87.2

80.6/80.6

83.9/82.3

89.2/85.6

81.4/75.0

82.7/82.7

83.4/83.4

94.1/90.2

82.0/82.0

82.0/79.5

Plan-Act

70.3/68.5

88.0/88.0

81.9/81.9

84.8/83.3

88.7/84.9

81.9/75.7

81.7/81.7

84.1/84.1

93.4/89.1

80.0/80.0

82.2/79.7

[2.5pt][2.5pt]

ConPAct-I

71.7/70.1

86.5/86.5

82.3/82.3

85.3/83.8

90.3/87.1

82.6/76.7

81.5/81.5

82.7/82.7

94.6/91.1

84.4/84.4

82.9/80.6

GPT-5.6-sol

ReAct

82.6/82.6

80.8/80.8

78.7/78.7

83.0/83.0

78.2/69.6

83.5/78.5

87.5/87.5

86.7/86.7

86.8/ 82.4

91.3/91.3

83.2/81.2

Plan-Act

84.8/84.8

76.9/76.9

78.7/78.7

85.1/85.1

77.3/69.6

83.6/78.5

83.3/83.3

86.7/86.7

87.1/ 82.4

87.0/87.0

82.9/80.9

[2.5pt][2.5pt]

ConPAct-I

82.6/82.6

76.9/76.9

80.9/80.9

85.1/85.1

81.7/73.9

85.9/80.7

83.3/83.3

86.7/86.7

88.2/82.4

91.3/91.3

84.1/82.0

Appendix

Table 17: Full OSWorld results, averaged over three rollouts per task. Each entry is Score/SR (%;

\uparrow

), rounded to one decimal place. Bold marks the best value for each metric within each model and category, including ties, using the values before rounding.

Configuration	Correction Trigger	✗	✓	✓	✓
	Conflict FeedBack	✗	✗	✓	✓
	Planner Participation	✗	✓	✗	✓
GPT-5.6-sol $\rightarrow$ GPT-5.6-sol
Sokoban	avg@8	85.0	90.8	89.8	92.1
	pass@8	96.0	97.5	97.0	98.0
	pass 8	71.0	81.5	78.0	82.5

Appendix

Table 18: Effects of online consistency ablations on task performance and planner–actor state mismatch.

Method	Sokoban			MiniGrid
	avg@8	pass@8	pass 8	SR	Return	Steps
GPT-5.6-sol
ReAct	83.8	97.0	59.5	54.5	0.50	10.9
Self-Refine	86.5	93.5	73.5	51.2	0.49	7.2
Plan-Act	79.4	91.5	52.5	53.1	0.46	13.1
Assertion	82.8	92.5	55.5	57.7	0.51	13.3

Appendix

Table 19: Inference controls on Sokoban and MiniGrid.

Environment	Metric	Baseline	No Sup.	No Replay	Full
Qwen3-VL-8B $\rightarrow$ Qwen3-VL-8B
Sokoban	$\mathrm{avg}@8$	40.7	44.4	43.5	49.1
	$\mathrm{pass}@8$	87.5	89.0	91.0	92.5
	$\mathrm{pass}^{8}$	6.5	9.0	5.5	11.0
	State mismatch	46.7	34.7	39.8	12.9
MiniGrid	SR	18.0	20.0	11.6	24.8

Appendix

Table 20: Training-data ablations on Sokoban and MiniGrid across the three planner–actor configurations. No Sup. and No Replay retain consistency filtering; Full uses all three components.

Environment	Metric	Baseline	ConPAct-R	$\Delta$
Qwen3-VL-8B $\rightarrow$ Qwen3-VL-8B
Sokoban	Initial D	47.4	20.8	$-26.6$
	Planner correctness	45.5	83.0	$+37.5$
	Actor correctness	46.6	83.1	$+36.5$
MiniGrid	Initial D	51.3	29.9	$-21.4$
	Planner correctness	50.9	81.5	$+30.6$

Appendix

Table 21: Initial state mismatch and assertion correctness before discussion. Rates are percentages and changes are in percentage points.

Training data		Sokoban			MiniGrid
Sup.	Replay	Discuss.	Init. D	Resid. D	Discuss.	Init. D	Resid. D
Qwen3-VL-8B
✗	✓	8.3	39.1	92.4	7.2	42.7	93.2
✓	✗	21.7	45.9	90.1	19.4	42.2	87.2
[][]✓	✓	58.4	20.8	62.0	54.1	29.9	78.9
Qwen3-VL-32B

Appendix

Table 22: Discussion responses and planner–actor state mismatch (%).

Training data		Sokoban			MiniGrid
Sup.	Replay	CC $\pm\Delta$	CW $\pm\Delta$	D $\pm\Delta$	CC $\pm\Delta$	CW $\pm\Delta$	D $\pm\Delta$
Qwen3-VL-8B
✗	✓	$40.7{+}1.5$	$19.8{+}1.5$	$39.5{-}3.0$	$38.4{+}1.6$	$17.9{+}1.5$	$43.7{-}3.1$
✓	✗	$28.4{+}2.4$	$28.0{+}2.0$	$43.6{-}4.4$	$36.2{+}3.3$	$17.1{+}2.7$	$46.7{-}6.0$
[][]✓	✓	$72.6{+}5.1$	$6.9{+}2.8$	$20.5{-}7.9$	$67.0{+}4.1$	$4.0{+}2.2$	$29.0{-}6.3$
Qwen3-VL-32B

Appendix

Table 23: State-assertion consistency and correctness: initial values and signed changes after discussion.

Panel of Figure 8 — Figure 8: Sokoban: state mismatch and ConPAct coordination. Condensed recorded outputs for the same positioning sub-goal.

Panel of Figure 9 — Figure 9: MiniGrid: state mismatch and ConPAct coordination. Condensed recorded outputs for the same gap-alignment sub-goal.

Panel of Figure 10 — Figure 10: Crafter: state mismatch and ConPAct coordination. Condensed recorded outputs for the same second-wood sub-goal.

Panel of Figure 11 — Figure 11: Procgen Miner: state mismatch and ConPAct coordination. Condensed recorded outputs for the same one-cell positioning sub-goal.

Method	Calls / episode	Output tokens / episode	Calls / action	Usage coverage (%)
ReAct (Planner)	14.02	3,095.37	1.04	71.5
ReAct (Actor)	36.29	8,703.19	1.00	71.5
Plan-Act	35.36	5,285.03	1.06	100.0
HiPlan	51.86	7,992.10	2.10	100.0
TAPE	37.84	18,900.17	6.29	99.9
[][] ConPAct-I	33.02	7,199.86	1.43	83.3

Appendix

Table 24: Inference costs averaged across the four environments and the evaluated inference-time configurations. Usage coverage is the fraction of recorded rollouts with output-token usage.

Explore similar work

Aug 3, 2026cs.RO

SyncPlan: Long-Horizon LLM Coordination with Explicit Synchronization and Adaptive Correction

LLM-based multi-agent coordination faces a fundamental trade-off between efficiency and adaptivity in dynamic environments. Existing approaches typically rely on repeated LLM invocations or multi-round communication to adapt decisions during execution, introducing substantial latency and making coordination vulnerable to asynchronous progress and environmental changes. Conversely, one-shot planning reduces coordination overhead but produces open-loop plans that can quickly become stale or fail when actions depend on other agents and the environment. We introduce SyncPlan, a plan-execute-correct framework for long-horizon coordination through explicit synchronization and adaptive correction. Given the state and team-level task, a centralized LLM coordinator generates per-agent action chains in a single planning call. During execution, explicit wait primitives and deadlock detection enforce inter-agent and agent-environment dependencies, while a lightweight Plan Staleness Detector continuously assesses the remaining plan and triggers replanning when environmental changes invalidate its assumptions. We further optimize the coordinator through SFT and planning-oriented RL with dense task progress and outcome-level execution feedback. Experiments on the public Overcooked benchmark and the complex Honor of Kings environment show that SyncPlan achieves state-of-the-art task success rates while using less than 0.05% of the wall-clock runtime compared with existing LLM-based coordinators. Code and datasets will be made publicly available.

Shen You, Xiaoming Zhu, Weining Weng +17

1City University of Hong Kong · 2Tencent · 3Shenzhen University +1

Coordination Gradient Staleness

Ask PDF

Sep 26, 2026cs.AI

Adaptive Consistency Graph for Long-Horizon Agents

Large language model agents can often make reasonable local decisions on short tasks, yet their performance degrades when success requires long sequences of dependent actions and tool calls. During execution, task requirements, historical evidence, and the current execution state may gradually become disconnected, so later decisions can drift from the original objective. We study this problem by introducing the Adaptive Consistency Graph (ACG) for long-horizon execution. ACG incrementally organizes execution evidence and its provenance in a persistent graph, then constructs a temporary requirement-centered view for each decision under a bounded context budget. Rather than replacing the base agent's planner or tool executor, ACG provides a structured and traceable context view for each decision. In the matched evaluation, ACG improves GPT-5.6-luna's average success from 44.5% with ReAct to 50.2%, with the largest gain on BrowseComp-Plus (73.5% versus 62.4%). We further analyze trajectory structure and inference cost to characterize this improvement. Our code is available at https://github.com/yunsaijc/Adaptive-Consistency-Graph.

Jiecong Wang, Hao Peng, Zhanyi Wang

Beihang University · QI-ANXIN GROUP

Long-Horizon Agents Long-Horizon Task Planning Agent Loop Large Language Model Agents Trajectory Generation Token Budget Allocation

Ask PDF

May 19, 2026cs.RO

ContextFlow: Hierarchical Task-State Alignment for Long-Horizon Embodied Agents

Long-horizon embodied agents increasingly delegate navigation, search, approach, and manipulation to specialist executors. As these executors become stronger, the main bottleneck shifts from local skill execution to maintaining a coherent task frontier across planning, monitoring, memory, and execution. We study task-state misalignment, a task-level consistency failure in which the planner's active stage, runtime evidence, remembered context, and delegated executor no longer justify the same next-step decision. This failure can lead to unsupported handoffs, stage lock, executor-context mismatch, and unnecessary replanning. We propose ContextFlow, an inspectable alignment framework that represents stages as explicit contracts, converts runtime observations into evidence packets, and applies scoped updates including continue, refine, transfer, promote, and repair. ContextFlow keeps specialist executors responsible for local closed-loop control while making task-frontier alignment explicit and auditable. Experiments and demonstration traces on long-horizon embodied tasks illustrate how evidence-grounded scoped updates diagnose and mitigate recurring task-state failures.

Shuhan Guo, Kun Zhang, Haifei Liu +4

Department of Electronic Engineering, Tsinghua University · Qiuzhen College, Tsinghua University · Institute of Microelectronics, Chinese Academy of Sciences +3

Long-Horizon Agents Embodied Agents Robot State

Ask PDF