Symbolic constraints improve ai agents tool call reliability
Symbolic Constraint Agents: Mining Formal Tool-Call Constraints to Prevent Agent Errors
Artificial Intelligence
Summary
Large language models (LLM) use tools via program calls, but they can make mistakes that cause failures. This paper shows that AI agents can learn rules about when and how to call these tools by analyzing manuals and past runs. The learned rules help find errors and fix them automatically, making the AI finish tasks more successfully. These improvements are stronger when errors are costly or hard to spot.
What this means in practice
- •For software developers: Improve the reliability of AI assistants by automatically mining and enforcing rules about tool usage from documentation and past behavior.
- •For automated customer support teams: Reduce failed AI tool interactions and improve task completion rates by embedding mined constraints to check and repair tool calls in AI systems.
Authors
Jinu Lee, Shraddha Barke
Abstract
Symbolic constraints (programmatic, deterministic checks for tool calls) enable deterministic verification of LLM agents' tool calls, but existing approaches primarily enforce predefined safety policies rather than tool dependencies in complex tool environments. We explore whether symbolic constraints can be mined from environments and observations beyond predefined policies, and whether they can improve end-to-end task performance. We present Symbolic Constraint Agents (SCA), a framework that mines symbolic constraints from API documentation, refines them against execution traces, and checks the agent's tool calls with an SMT solver at runtime. A violation triggers a natural-language warning or, when the valid replacement is uniquely determined, an automatic repair. On AppWorld, refinement raises error-detection F1 from 37.0 to 58.3 on GPT-5.5 traces and from 39.3 to 61.9 on Qwen3.5-27B traces. At runtime, repair raises GPT-5.4-mini's task success by 5.4 points under a 40-call budget and 9.8 points when environment errors are hidden, while removing 59--87\% of failed calls across models, and transfers to $τ^2$-bench with gains of up to 5.6 points. These findings show that automatically acquired symbolic constraints can improve tool-calling reliability and task completion, particularly when failures are costly or difficult to diagnose.