Large language model (LLM) agents acting across web applications and online services often need multi-turn corrections to align their work with users' underspecified goals and preferences. In practice, users may identify problems in the current outcome without specifying the desired result, so learning user intentions requires interpreting each correction against the agent's preceding and revised outcomes. These expectations are often connected to a particular context, making it hard to know when they apply and how to ensure the agent later acts as the user expects. To this end, we propose LiveAsset, an intent harness that learns reusable intent assets from corrective interactions and uses them for runtime control of agent execution. LiveAsset jointly interprets each correction with the agent's original and revised results to infer user intent, incrementally updating the assets while preserving earlier requirements that remain applicable. The learned requirements guide agent execution from task entry and provide criteria for runtime verification, which uses the explanations to assess applicability before each reply is delivered and directs the agent to revise the result to address unmet requirements while preserving those already satisfied. Following a learning phase in which simulated users progressively reveal requirements through multi-turn corrections, we evaluate LiveAsset using same-task replay of the original requests on π-Bench, PERMA, and Vita and cross-task transfer on PAST-Bench and SkillEvolBench. Experiments show that LiveAsset better fulfills requirements learned from corrective interactions and improves performance on these benchmarks compared with memory, skill, and rule-based harness baselines. Code is available at github.com/landian60/LiveAsset.
Figures & tables
Figure 1. Challenges for existing approaches in learning from corrective interactions in human–agent collaboration. The top panels report (a) requirement satisfaction with stored skills, (b) retrieval coverage of memory records written after corrections, and (c) requirement fulfillment after rule-based release, and the bottom panels illustrate the corresponding failures.
Figure 2. Feedback forms in human–agent collaboration, analyzed to inform the design of simulated corrective interactions. Stated requirements (left) and corrective feedback (right) are counted over 331 SWE-chat messages, with excerpts showing their forms and overlap.
Figure 3. Overview of LiveAsset. Left: incremental learning of intent assets from corrections and the original and revised results. Right: situation matching selects relevant intents, and runtime checks assess their applicability and fulfillment to guide revision by the same agent.
Same-task replay
Transfer to changed tasks
π -Bench
PERMA
Vita
PAST-Bench
SkillEvolBench
Method
Satisfaction
PROC
Satisfaction
TC
Satisfaction
Reward
All
Near
Far
All
T4
T5
T6
Actor: GPT-6 Luna
Base agent
0.218
0.466
0.132
0.114
0.228
0.015
0.641
0.648
0.634
0.815
0.809
0.840
0.794
Raw history
0.801
0.794
0.881
0.929
0.789
0.424
0.933
0.950
0.910
0.857
0.892
0.858
0.821
Skill
0.554
0.681
0.704
0.657
0.733
0.545
0.836
0.839
0.833
0.854
0.882
0.855
0.826
Table 1. Performance on same-task replay and transfer across five benchmarks and two actor models. Satisfaction measures fulfillment of exposed user requirements; PROC incorporates delivery evidence, while TC and Reward measure task completion and trajectory quality. Bold marks the best result among Skill, Mem0, TellOnce, and LiveAsset.
Figure 4. Situation selection in same-task replay. (a) Top-1 selections on Vita, counted over two retrieval runs. (b, c) Recall@ k of each task’s learning histories on PERMA and Vita.
Figure 5. Runtime verification and requirement fulfillment on π -Bench. (a) External satisfaction of the first candidate and final delivery. (b) Final runs that pass runtime checks but leave requirements unmet. (c) Paths of initially unmet requirements, shown as proportions of each method’s n .
Method / variant
DeepSeek V4.1-Flash
GPT-6 Luna
Sat.
Task score
Sat.
Task score
π -Bench (PROC)
LiveAsset (Full)
0.921
0.964
0.925
0.969
w/o explanation
0.827
0.833
0.788
0.912
w/o context
0.862
0.852
0.881
0.908
w/o runtime
0.912
0.915
0.883
0.924
Table 2. Ablation results for the full system and three variants on three replay benchmarks.
Figure 6. Requirement fulfillment across repeated runs on π -Bench. (a) Per-task satisfaction. (b) Absolute score difference between two runs. (c) Per-task check outcomes, with the share of checks passed twice and the number of tasks passing all checks twice.
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
Benchmark
Mˉ
Cˉ
π -Bench
2.90
0.70
PERMA
2.11
2.09
Vita
4.27
0.85
PAST-Bench
1.83
1.83
SkillEvolBench
1.53
1.53
Appendix
Table 3. Mean user follow-ups ( Mˉ ) and corrective messages ( Cˉ ) per learning task or episode across five benchmarks.
Benchmark
Learning and evaluation scope
Metric
π -Bench
100 tasks, 5 personas
Exposed checklist satisfaction; PROC with delivery evidence.
6 environments, 30 families, 180 tasks; T1–T3 learning and T4–T6 testing, 90 each
Native verifier reward and process/outcome checks.
Appendix
Table 4. Benchmark coverage and evaluation measures for Table 1 . Both actor models and all compared methods use the full evaluation scope listed below. Counts exclude repeated executions.
DeepSeek V4.1-Flash
GPT-6 Luna
Method
Test-time (k/run)
Build (k/task)
Test-time (k/run)
Build (k/task)
LiveAsset
367.1
72.7
205.9
66.3
Base agent
243.6
–
154.2
–
Raw history
276.9
–
293.0
–
Skill
410.9
380.8
170.6
259.7
Mem0
308.0
99.4
155.6
107.0
Appendix
Table 5. Mean token use for six methods across π -Bench, PERMA, and Vita, in thousands of tokens per execution (Test-time) or learning task (Build).
Figure 7. Intent checks and revisions across two runs per task. Triangles mark continuation, with numerals for multiple revisions; purple outlines mark different revision counts. Crosses below final checks mark external failure; unmarked runs pass. External criteria follow Appendix D.3 .