Robust Function Calling for LLM Agents Under Interface Drift
SchemaShield-Lite restores Qwen3.5-9B function-calling accuracy after tool-interface drift without retraining.
Researchers study interface drift, where tool descriptions, argument schemas, or candidate lists change after deployment while tool semantics stay fixed. Their training-free SchemaShield-Lite layer combines schema validation, candidate-set retry, and one-shot constrained repair. On Qwen3.5-9B it lifts DICE-300 accuracy from 0.717 to 0.813 and BFCL-200 from 0.580 to 0.805, recovering 19/22 and 33/45 drift-induced failures. Ablations show candidate-set drift, especially distractor tools, causes most of the damage, and naive retry does not explain the gains.
- SchemaShield-Lite validates schemas, retries missing tools, and does one-shot constrained repair.
- On Qwen3.5-9B, DICE-300 rose from 0.717 to 0.813 and BFCL-200 from 0.580 to 0.805.
- It recovered 19 of 22 and 33 of 45 drift failures with no observed repair harms.
- Distractor tools in the candidate set caused most damage; description and schema edits were milder.
Full article153 words · extracted from arxiv.org · click to collapse
Tool-augmented language-model agents rely on interface text: tool descriptions, argument schemas, and the candidate tool list shown at inference time. These interfaces often change after deployment even when the underlying tool semantics remain fixed, creating a practical reliability problem for function-calling systems. We study this failure mode as interface drift and evaluate whether a lightweight, training-free repair layer can recover robustness without retraining the base model. Our method, SchemaShield-Lite, combines schema validation, candidate-set retry for missing tool names, and one-shot constrained repair. On Qwen3.5-9B, it raises DICE-300 accuracy from 0.717 to 0.813 and BFCL-200 accuracy from 0.580 to 0.805, recovering 19/22 and 33/45 drift-induced failures with no observed repair harms. Naive retry does not explain the gains. Ablations show that candidate-set drift, especially distractor tools, accounts for most observed damage, while description and schema edits alone are less harmful. The results position validation-and-repair as a practical robustness layer for deployed LLM-agent tool use.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2610.23342