Large language model (LLM) agents provide a promising interface for command-line-based network operations, but a plausible command may still fail or introduce operational risk after execution. Existing approaches mainly focus on command generation or final configuration correctness, and do not use execution-grounded experience to jointly improve candidate coverage and action selection. We propose an execution-grounded dual-path consequence-aware agent for CLI-based SONiC operations, which generates multiple complete actions, predicts their execution consequences, and selects the final action through utility- and risk-aware reranking. The proposal-side path abstracts reusable operational lessons into retrievable guidance to improve feasible-action coverage without modifying the proposal LLM, while the selection-side path adapts the consequence predictor through session-level LoRA updates using real SSH feedback to improve conditional selection quality. Experiments over multi-turn SONiC operation sessions with different Qwen3 proposal models show that the framework improves feasible-action coverage and top-1 execution success, and that the two adaptation paths provide complementary gains over interaction.
Good action rankings do not make a contrastive critic safe to maximize. These critics increasingly act as value-like objectives for best-of-$K$ selection, planning, and critic-guided generation. Unbounded bilinear scores can let large embedding norms inflate off-support values, but cosine bounding does not remove the failure. A controlled support decomposition attributes most raw bilinear regret to norm drift. Cosine and hybrid critics nevertheless select off-support actions from most pools and incur comparable regret. Contrastive scores are weakly calibrated or inverted in the top score decile across four OGBench navigation tasks, and they fail to order fixed-query actions by value. Bellman-trained TD-Q succeeds, including in a parameter-matched function-class control. Realized costs depend on the task: simulator rollouts reveal single-step selection costs on PointMaze and the exact-$Q^*$ toy but well-powered nulls on AntMaze and HumanoidMaze, where the controller can self-correct. A training/readout decomposition traces the lost ordering to the cosine training objective; raw-trained embeddings retain weak ordering after inference-time normalization. Candidate maximization can therefore exploit false positives caused by norm drift, score saturation, or in-support misranking. Contrastive critics remain useful compatibility rankers on navigation and manipulation tasks, but action selection requires a value-calibrated scalar.
Liangyu Li, Qingwen Liu, Mingqing Liu +1cs.RO cs.AI cs.LG
Controllers based on sampling and latent world models assign a predicted terminal cost to each candidate action sequence, choose the minimum, execute its first action block, and replan. This rule can fail even when the terminal cost perfectly and accurately reflects the true task objective in the physical world. Residual prediction error can give an infeasible sequence an anomalously low cost, and a larger proposal pool gives such errors more chances to outrank feasible alternatives. We call this conditional failure proposal overgeneration. In Cube candidate execution audits, increasing the total proposal budget from 72 to 288 reduces the feasibility of selection by minimum latent cost from .375 to .062 for position targets and from .344 to .031 for targets defined by position and yaw, although every larger pool contains a feasible sequence. We introduce Adjacent Set Action Reconstruction (ASAR). Among proposals with low cost, ASAR identifies an adjacent set using standardized early action prefixes and reconstructs a full action sequence through locally weighted aggregation with a light anchor from the sequence with minimum cost. On a Carry and Release evaluation set of 75 queries, Kernel ASAR improves event completion success over matching selection by 28.0, 24.0, and 18.7 percentage points under latent cost and by 18.7, 20.0, and 17.3 points under a trajectory reachability cost at 72, 144, and 288 proposals. Analysis of finite proposal pools characterizes selection risk from the lower tail, separation by a related radius support statistic, and sequence containment under an explicit local feasibility condition.
LLM agents choose tools and arguments from context that mixes user requests, tool outputs, retrieved records, memory, and untrusted text. Evidence can be relevant without being authorized to determine a decision, so a correct action need not be grounded only in permitted evidence. We introduce a target-specific authorization audit that labels context factors separately for each tool and argument target. Its primary test holds the task, proposition, position, and policy fixed while changing only the proposition's source authority. We then test behavior when valid evidence is weakened and use context-subset interactions as a secondary localization diagnostic. Across 450 controlled next-action tasks and multiple open-weight LLM families, trusted and untrusted variants produce different actions in 5.4 percent of competing cases versus 1.7 percent of supporting cases. Under controlled degradation, unauthorized competition is retained in a full-correct, mixed-error, clean-correct pattern in 2.4 percent of comparisons, with a 95 percent confidence interval from 2.1 to 3.0 percent. These are controlled stress-set rates, not deployment prevalence. The models respond to textual source-authority cues, but this does not prevent untrusted evidence from influencing their actions.
Jihyeok Jung, Jeewu Lee, Sanghyeop Kim +2cs.CV cs.AI
Existing egocentric benchmarks have primarily constructed the egocentric setting from first-person-view data, which makes it difficult to evaluate egocentric perspective itself in isolation. However, understanding first-person-view input and taking an egocentric perspective are separable abilities, especially when first-person body cues are absent or when other agents are present. To isolate egocentric perspective understanding, we introduce EgoGapBench, a diagnostic benchmark for measuring action selection in multi-agent egocentric scenes. We define the ability measured by this benchmark as Egocentric Action Selection (EAS): selecting an appropriate action from the agent's perspective in the presence of other agents. On EgoGapBench, humans answer reliably, whereas both open-source and proprietary MLLMs perform substantially worse and systematically select actions performed by other visible agents. Fine-tuning on existing egocentric data fails to close this gap and can even be detrimental. In contrast, fine-tuning on EgoGapBench training data improves accuracy but does not reach human performance. These results show that EAS is difficult to acquire from first-person-view data alone, and that MLLMs should be evaluated and trained not only for scene understanding but also for egocentric action selection.
Large language model agents achieve strong performance on text-based benchmarks but incur prohibitive inference costs, motivating the use of compact neural rerankers for action selection. We investigate whether a single lightweight model can perform action selection across multiple diverse environments, a capability that would eliminate per-environment model maintenance. Training DeBERTa-v3 (184M-434M parameters) jointly on ALFWorld, WebShop, and ScienceWorld with minority-class upsampling, we find that rebalanced two-environment joint training substantially improves over single-environment ALFWorld performance (net gain +0.412) while maintaining competitive WebShop performance (+0.214 vs. +0.249 single-environment). Three-environment training yields a mean combined net gain of +0.551 +/- 0.024 across 4 seeds, with per-environment results approaching specialized single-environment models while providing positive cross-domain transfer. Cross-environment adaptation is highly sample-efficient: fine-tuning on only 9.2% of target-domain data recovers 93% of full-data performance, and scaling model capacity yields limited benefits, indicating data diversity is the primary driver. Environment-aware LoRA adapter routing with PCGrad achieves a best-seed result of +0.611 (seed 42), with seeds 456 and 789 at +0.554 and +0.559, but exhibits high variance due to seed 123 collapsing to +0.263 (4-seed mean +0.497 +/- 0.158), representing a promising but currently unstable direction. Joint training with clean splits and data rebalancing is a key ingredient. We will release our three-environment benchmark of 51,580 training instances (41,740 raw unique states with minority-class upsampling) and all model checkpoints upon acceptance.