Measuring the Serving Stack Instead of the Model: Hidden Confounds in Local Tool-Use Evaluation
Read the original on arXiv Computation and Language →The Flow has not summarised this story yet — read it at arXiv Computation and Language.
The Flow has not summarised this story yet — read it at arXiv Computation and Language.
arXiv:2608.29128v1 Announce Type: new Abstract: Tool-using agents are commonly evaluated by a single bit: whether an end-to-end workflow completed. This metric fails to distinguish failures that matt...
The paper introduces prefix executability, a trajectory-level evaluation protocol for tool-using agents that checks whether every prefix of a predicted tool-call sequence remains executable under tool schemas, state preconditions, and data dependencies. It defines metrics such as Prefix Executability Curves, First-Error Depth, Trajectory Executability Rate, horizon-normalized AUC-PEC, and Recovery Rate to quantify reliability. Experiments on BFCL-v3, ToolSandbox, and $ au^3$-bench show that agents can achieve high final success while having low trajectory executability, revealing failure modes that final success metrics miss, and demonstrate that explicit pre-execution plan checking can improve prefix-level reliability without sacrificing terminal success.
The paper introduces isloss, a deterministic oracle that tests whether conversations are preserved accurately when translated across different LLM APIs. It maps source and translated requests and replies to a shared intermediate representation, ConvIR, and applies explicit rules to detect silent information loss such as changes to instruction roles, tool interactions, error states, or generation controls. Evaluation shows that isloss can detect a majority of planted faults with few false alarms and reveals that many translations silently alter critical conversation details, affecting model behavior.
The paper "Log-To-Leak: Prompt Injection Attacks on Tool-Using LLM Agents via the Model Context Protocol" investigates how natural-language metadata in the Model Context Protocol (MCP) can be exploited to inject malicious post‑hoc logging calls into large language model agents. The authors introduce the Log-To-Leak attack, which adds an attacker‑controlled call after a legitimate tool invocation, targeting user queries, MCP server names, server responses, and agent responses. Experiments across five MCP servers and four LLM agents show high attack success rates (up to 100%) and significant logging accuracy, while existing defenses still leave substantial residual risk.
arXiv:2609.14758v1 Announce Type: cross Abstract: Tool-augmented language models are evaluated on whether they reach the right answer, not on whether they report honestly when a tool fails to supply...
UniACE is a unified framework that standardizes the evaluation of large language model (LLM) agents by representing each benchmark as an instruction–tool–environment triplet and running models through a shared, task‑agnostic harness in isolated runtimes. It preserves native success criteria, offers an offline mode for dynamic‑resource tasks, and standardizes efficiency metrics, execution records, and failure attribution. Applying UniACE to 7 benchmarks across 24 domains and 15 models revealed significant score shifts, ranking reversals, and sensitivity to evidence representation, highlighting the impact of evaluation configuration on reported agent performance.