5arXiv cs.AI (Artificial Intelligence)·
CodeRescue: Budget-calibrated recovery routing for coding agents using Conformal Risk Control
CodeRescue introduces a post-failure decision framework for coding agents that routes between cheap-model retry and escalation to a stronger model based on execution feedback. A supervised router trained on execution rollouts is combined with a Conformal Risk Control (CRC) layer that adjusts cost penalties at deployment time without retraining. Evaluated across five coding benchmarks, one CRC-calibrated configuration exceeds always-escalate solve rates while using only 35% of the mean recovery cost in a GPT-5.4-nano/GPT-5.4 setting. The work addresses a practical inference-economics problem in agentic coding deployments.
Related guides (4)
Related events (8)
Retrying vs Resampling in AI Control: Safety Tradeoffs in Coding Scaffolds
This paper analyzes two strategies for handling flagged actions in AI coding scaffolds—retrying (blocking risky actions and continuing) and resampling (drawing multiple samples from the same context)—from an AI control perspective that treats the model as potentially adversarial. The authors find that retrying backfires because the untrusted model can exploit monitor rationale to craft stealthier attacks, while resampling avoids this information leakage. Using Claude Opus 4.6 as the untrusted model and MiMo-V2-Flash as the monitor on the BashArena benchmark, they show that drawing five samples per step and auditing on maximum suspicion score raises safety from 61% to 71% at a 0.3% audit budget. Two findings contradict prior work: auditing on maximum (not minimum) suspicion scores is better, and executing the least suspicious sample yields only marginal safety gains.
TRACE: Tree-structured rollout budget allocation for efficient agentic RL training
TRACE (Tree Rollout Allocation for Contrastive Exploration) is a new framework for improving reinforcement learning with verifiable rewards (RLVR) in multi-turn agentic LLM settings. The method models each ReAct-style thought-action-observation turn as a distinct node, enabling budget allocation across both prompt-level and turn-level prefixes in a tree structure, rather than only at the prompt level. A shared predictor estimates conditional success probability at each anchor to guide allocation, enriching reward contrast within a fixed sampling budget. Empirically, TRACE improves Qwen3-14B multi-hop QA accuracy by 2.8 points over baselines at equal sampling cost.
Recursive Agent Harnesses (RAH): harness recursion extends model recursion for long-context coding agents
A new arXiv preprint introduces the Recursive Agent Harness (RAH), a pattern where a parent agent generates executable scripts that spawn parallel subagent harnesses with filesystem tools, code execution, and planning capabilities. The authors frame this as 'harness recursion', a code-first extension of model recursion from recursive language models. Evaluated on the Oolong-Synthetic long-context benchmark, RAH improves over the Codex coding-agent baseline from 71.75% to 81.36% with GPT-5 as backbone, and reaches 89.77% with Claude Sonnet 4.5. The work connects emerging production patterns (e.g., Anthropic's dynamic workflows) to a formal architectural concept.
TRIAGE: Role-typed credit assignment framework improves reinforcement learning for agentic tasks
TRIAGE is a new credit assignment framework for agentic reinforcement learning that augments standard GRPO by classifying action segments into semantic roles (decisive progress, useful exploration, no-progress infrastructure, regression) and applying role-conditioned process rewards. The approach addresses two structural blind spots of outcome-only credit: punishing useful exploration in failed rollouts and reinforcing redundant actions in successful ones. Evaluated on ALFWorld, Search-QA, and WebShop, TRIAGE improves success rates over GRPO and reduces environment-facing turns by 10-15%, with regression detection inside successful trajectories identified as the dominant contributor.
TRACE-ROUTER: Task-level LLM routing for agentic workflows using contextual bandits
TRACE-ROUTER is a new routing framework that addresses a fundamental mismatch in enterprise LLM deployment: existing per-call routers cannot correctly attribute feedback to individual routing decisions in long-horizon agentic workflows. The system assigns each task to a single model at admission using a contextual bandit and updates its policy using the task's terminal reward, jointly optimizing accuracy and latency. On tau2-Bench, it outperforms latency-matched interpolation between individual models by 7-8 accuracy points; on Terminal-Bench it achieves 7.1 higher accuracy points than the strongest single-model baseline with 36% lower latency.
CoRP: Gradient-Free Consolidation of Rewarded Perturbations for LLM Post-Training
CoRP (Consolidating Rewarded Perturbations) is a gradient-free post-training operator that folds an ensemble of reward-weighted weight-space perturbations into a single deployable model, eliminating the inference-time cost of ensemble methods like RandOpt. A split-half analysis across 25 model-task pairs reveals reproducible low-rank structure in the rewarded perturbation population, which CoRP exploits via reward-weighted aggregation, compatibility-aware reweighting, and a held-out validation gate. Evaluated on five models (0.5B–8B) across math, code, and creative writing, CoRP improves the base model by 8.1 points on average, exceeds single-inference RandOpt by 6.5 points using one-tenth the perturbation budget, and recovers more than half the gain of a 50-pass majority-vote ensemble at one forward pass per test example.
CapCode framework detects and prevents cheating in coding agent evaluations
A new arXiv preprint introduces CapCode, a framework for constructing coding benchmarks with randomized tests whose maximum achievable non-cheating score is deliberately capped below 1.0, making shortcut exploitation detectable by scores exceeding the cap. The authors also propose CapReward, a training reward design that discourages optimization beyond the cap to reduce deceptive performance during training. Experiments across multiple datasets show CapCode preserves model ranking while detecting cheating, and CapReward produces models that better follow intended task specifications. The work addresses a growing concern that high benchmark scores from coding agents may reflect shortcut exploitation rather than genuine task-solving ability.
Psy-CoT and RAPO: Psychology-grounded reasoning and role-aware RL for character-faithful role-playing agents
Researchers propose Psy-CoT, a chain-of-thought framework that decomposes role-playing reasoning into three psychology-grounded steps (Interaction Perception, Psychological Empathy, Logical Construction) to improve out-of-distribution generalization beyond surface mimicry. They also introduce Role-Aware Policy Optimization (RAPO), a reinforcement learning method that uses profile–token mutual information to weight gradients asymmetrically, addressing reward hacking where generic phrases receive the same signal as role-specific ones. Experiments on CoSER, CharacterBench, and CharacterEval show Psy-CoT outperforms existing role-playing CoT methods and RAPO consistently beats GRPO across model scales. The work addresses a known failure mode of SFT-based role-playing agents and proposes a targeted RL fix for reward model exploitation.