LogDx-CI: Benchmarking Log Reduction Tools for LLM Root-Cause Diagnosis

2026-05-29T08:52:14Z1f26202b3ee2859090842a51e330cc399ec20bd0b9ee392736bea50cd6984a57
agent-loopbenchmarkingci-logscode-conversiondecompilationllm-in-the-looplog-reductionobservational-equivalencepatch-repairprompt-injectionreproducibilitysecurity-evaluationsmart-contractvulnerability-dataset

What happened

This collection contains several 2026 arXiv papers with security-relevant findings for LLM-integrated systems and code/tooling: (1) LLMCVE: a curated dataset of 2,888 multi-source vulnerabilities across 230 popular LLM components; manual triage identifies 205 LLM-in-the-loop vulnerabilities and shows LLMs are often targets/propagation vectors rather than root causes. Repair experiments show agent-based fixes (e.g., SWE-Agent) struggle—prompt-injection cases have a Pass@1 of only 28.57%, indicating repairs are challenging. (2) SCDBench: a 600-contract benchmark and methodology for LLM-based de-

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arxiv_cs_se
Record identifier
1f26202b3ee2859090842a51e330cc399ec20bd0b9ee392736bea50cd6984a57
Enrichment time
2026-05-29T08:52:14Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.