SmartSnap: Proactive Evidence Seeking for Self-Verifying Agents

TL;DR

SmartSnap tackles the verification bottleneck in agentic RL for GUI tasks by introducing Self-Verifying Agents that proactively curate a minimal set of decisive evidences. Guided by the 3C Principles (Completeness, Conciseness, Creativity), agents perform in-situ self-verification and receive dense reward signals from a structured LLM/VLM verifier, enabling scalable training with GRPO. Empirical results on AndroidLab show consistent improvements across model scales (up to $26.08\%$ SR gains) and demonstrate that evidence-centric verification can rival larger baselines while reducing verification cost. The approach promises more robust, scalable autonomous GUI agents and sets the stage for broader, real-world deployment with diversified benchmarks and continual pre-training.

Abstract

Agentic reinforcement learning (RL) holds great promise for the development of autonomous agents under complex GUI tasks, but its scalability remains severely hampered by the verification of task completion. Existing task verification is treated as a passive, post-hoc process: a verifier (i.e., rule-based scoring script, reward or critic model, and LLM-as-a-Judge) analyzes the agent's entire interaction trajectory to determine if the agent succeeds. Such processing of verbose context that contains irrelevant, noisy history poses challenges to the verification protocols and therefore leads to prohibitive cost and low reliability. To overcome this bottleneck, we propose SmartSnap, a paradigm shift from this passive, post-hoc verification to proactive, in-situ self-verification by the agent itself. We introduce the Self-Verifying Agent, a new type of agent designed with dual missions: to not only complete a task but also to prove its accomplishment with curated snapshot evidences. Guided by our proposed 3C Principles (Completeness, Conciseness, and Creativity), the agent leverages its accessibility to the online environment to perform self-verification on a minimal, decisive set of snapshots. Such evidences are provided as the sole materials for a general LLM-as-a-Judge verifier to determine their validity and relevance. Experiments on mobile tasks across model families and scales demonstrate that our SmartSnap paradigm allows training LLM-driven agents in a scalable manner, bringing performance gains up to 26.08% and 16.66% respectively to 8B and 30B models. The synergizing between solution finding and evidence seeking facilitates the cultivation of efficient, self-verifying agents with competitive performance against DeepSeek V3.1 and Qwen3-235B-A22B.

Figures (13)

• Figure 1: Success rate on AndroidLab xu2024androidlab across model families and scales. Compared with the vanilla prompting (PT), our SmartSnap brings significant gains via fine-tuning (FT) and reinforcement learning (RL) without relying on sophisticated rule-based verifiers and task-specific reward models. The developed self-verifying agents learn to complete tasks and curate snapshot evidences in a complementary manner, achieving competitive performance with larger LLMs.
• Figure 2: Three strategies for agent verification distinguished by their inputs to the verifier: (a) the task-specific script accessing the ground-truth state; (b) the full trajectory with noisy context; and (c) the agent-curated evidence set.
• Figure 3: An example of self-verification for evidence curation. The agent decomposes the task into an actionable checklist where the date, the amount, and the category tag are to be confirmed during stepwise task completion. The proactive step of taking snapshots that list the target transaction provides a definitive evidence for task completion.
• Figure 4: Agent behavior evolution (RL dynamics).
• Figure 5: Agent performance variation (RL dynamics).