26.1% of analyzed AI agent skills contain vulnerabilities across 14 patterns, with executable scripts raising risk 2.12x, based on static and LLM analysis of 31k skills.
hub Mixed citations
Richard Landis and Gary G
Mixed citation behavior. Most common role is background (67%).
17
Pith papers citing it
Background
67% of classified citations
hub tools
citation-role summary
background 8
method 1
citation-polarity summary
years
2026 17representative citing papers
Urban-ImageNet is a 2-million-image multi-modal dataset with HUSIC 10-class taxonomy enabling benchmarks for urban scene classification, cross-modal retrieval, and instance segmentation.
VulKey reaches 31.5% repair accuracy on real C/C++ vulnerabilities by matching hierarchical expert patterns to guide LLM patch generation, beating prior baselines by 7.6%.
AgentProp-Bench shows substring judging agrees with humans at kappa=0.049, LLM ensemble at 0.432, bad-parameter injection propagates with ~0.62 probability, rejection and recovery are independent, and a runtime fix cuts hallucinations 23pp on GPT-4o-mini but not Gemini-2.0-Flash.
The C-Score quantifies intra-class explanation consistency for CAM methods via confidence-weighted pairwise soft IoU and detects AUC-consistency dissociation as an early warning for model instability on chest X-ray classification.
Participatory design with 20 Afghan women reveals that safe GenAI learning companions must prioritize privacy, cultural fit, and genuine learning support, with the process itself linked to higher aspirations and agency.
Analysis of SATD in Dockerfiles shows 27% of admissions and 40% of repayments are coupled to non-Dockerfile artifacts, with coupled events repaid faster overall and external dependencies as a key trigger.
An extended annotation scheme with new categories and attributes plus a Gemma-300M-based multi-head classifier achieves 81.6% macro F1 on personal fact classification, outperforming few-shot LLM baselines by nearly 9 points with lower compute.
Universal adversarial attacks cause output perturbation 90 times more often than precise target injection in VLMs, with only 2 verbatim successes out of 6615 tests.
A paraphrase-robust duplicate-step detector for Gherkin BDD suites, built on a new 1.1M-step public corpus, reports F1 scores up to 0.906 and estimates 893k eliminable step occurrences corpus-wide.
LLMs can detect usability content in user reviews with F-scores comparable to humans, though performance depends strongly on prompt design.
Larger batch sizes for LLM dialogue coding in healthcare simulations improve speed and reduce energy consumption while decreasing coding accuracy compared to human labels.
Decomposing BP annotation into 14 skills shows 5 directly operable, 4 recoverable after re-annotation, and 5 structurally underspecified, with GPT-5.4 reaching 0.678 accuracy on retained skills and human-GPT difficulty correlating at r=0.881 at the skill level but near zero at instance and lexical-1
ToxiShield delivers a real-time GitHub extension with a BERT toxicity detector at 98% accuracy, a Claude-based coach, and a fine-tuned Llama reframer at 95% style transfer accuracy, validated by a 10-person TAM study.
An automated self-testing framework with evidence-based quality gates for LLM application releases was evaluated in a longitudinal case study of a multi-agent conversational AI system, identifying rollback builds and supporting stable quality over four weeks.
Guided blog posts during work-based learning enable CS students to produce deep reflections on problem-solving, collaboration, and personal growth that they can use in resumes and interviews.
Verbalized confidence from small LMs enables cost-effective cascade routing for automated educational scoring, matching large-model accuracy at 76% lower cost when discrimination is strong.
citing papers explorer
-
Agent Skills in the Wild: An Empirical Study of Security Vulnerabilities at Scale
26.1% of analyzed AI agent skills contain vulnerabilities across 14 patterns, with executable scripts raising risk 2.12x, based on static and LLM analysis of 31k skills.
-
Urban-ImageNet: A Large-Scale Multi-Modal Dataset and Evaluation Framework for Urban Space Perception
Urban-ImageNet is a 2-million-image multi-modal dataset with HUSIC 10-class taxonomy enabling benchmarks for urban scene classification, cross-modal retrieval, and instance segmentation.
-
VulKey: Automated Vulnerability Repair Guided by Domain-Specific Repair Patterns
VulKey reaches 31.5% repair accuracy on real C/C++ vulnerabilities by matching hierarchical expert patterns to guide LLM patch generation, beating prior baselines by 7.6%.
-
Evaluating Tool-Using Language Agents: Judge Reliability, Propagation Cascades, and Runtime Mitigation in AgentProp-Bench
AgentProp-Bench shows substring judging agrees with humans at kappa=0.049, LLM ensemble at 0.432, bad-parameter injection propagates with ~0.62 probability, rejection and recovery are independent, and a runtime fix cuts hallucinations 23pp on GPT-4o-mini but not Gemini-2.0-Flash.
-
Quantifying Explanation Consistency: The C-Score Metric for CAM-Based Explainability in Medical Image Classification
The C-Score quantifies intra-class explanation consistency for CAM methods via confidence-weighted pairwise soft IoU and detects AUC-consistency dissociation as an early warning for model instability on chest X-ray classification.
-
Designing Safe and Accountable GenAI as a Learning Companion with Women Banned from Formal Education
Participatory design with 20 Afghan women reveals that safe GenAI learning companions must prioritize privacy, cultural fit, and genuine learning support, with the process itself linked to higher aspirations and agency.
-
Beyond the Tip of the Iceberg: Understanding SATD in Dockerfiles through the Lens of Co-evolution
Analysis of SATD in Dockerfiles shows 27% of admissions and 40% of repayments are coupled to non-Dockerfile artifacts, with coupled events repaid faster overall and external dependencies as a key trigger.
-
An Annotation Scheme and Classifier for Personal Facts in Dialogue
An extended annotation scheme with new categories and attributes plus a Gemma-300M-based multi-head classifier achieves 81.6% macro F1 on personal fact classification, outperforming few-shot LLM baselines by nearly 9 points with lower compute.
-
VisInject: Disruption != Injection -- A Dual-Dimension Evaluation of Universal Adversarial Attacks on Vision-Language Models
Universal adversarial attacks cause output perturbation 90 times more often than precise target injection in VLMs, with only 2 verbatim successes out of 6615 tests.
-
Reducing Maintenance Burden in Behaviour-Driven Development: A Paraphrase-Robust Duplicate-Step Detector with a 1.1M-Step Open Benchmark
A paraphrase-robust duplicate-step detector for Gherkin BDD suites, built on a new 1.1M-step public corpus, reports F1 scores up to 0.906 and estimates 893k eliminable step occurrences corpus-wide.
-
User Reviews as a Source for Usability Requirements: A Precursor Study on Using Large Language Models
LLMs can detect usability content in user reviews with F-scores comparable to humans, though performance depends strongly on prompt design.
-
Scalable LLM-based Coding of Dialogue in Healthcare Simulation: Balancing Coding Performance, Processing Time, and Environmental Impact
Larger batch sizes for LLM dialogue coding in healthcare simulations improve speed and reduce energy consumption while decreasing coding accuracy compared to human labels.
-
Exploring and Testing Skill-Based Behavioral Profile Annotation: Human Operability and LLM Feasibility under Schema-Guided Execution
Decomposing BP annotation into 14 skills shows 5 directly operable, 4 recoverable after re-annotation, and 5 structurally underspecified, with GPT-5.4 reaching 0.678 accuracy on retained skills and human-GPT difficulty correlating at r=0.881 at the skill level but near zero at instance and lexical-1
-
ToxiShield: Promoting Inclusive Developer Communication through Real-Time Toxicity Filtering
ToxiShield delivers a real-time GitHub extension with a BERT toxicity detector at 98% accuracy, a Claude-based coach, and a fine-tuned Llama reframer at 95% style transfer accuracy, validated by a 10-person TAM study.
-
Automated Self-Testing as a Quality Gate: Evidence-Driven Release Management for LLM Applications
An automated self-testing framework with evidence-based quality gates for LLM application releases was evaluated in a longitudinal case study of a multi-agent conversational AI system, identifying rollback builds and supporting stable quality over four weeks.
-
Writing Blog Posts Helps Students Connect Experiential Learning to the Workplace
Guided blog posts during work-based learning enable CS students to produce deep reflections on problem-solving, collaboration, and personal growth that they can use in resumes and interviews.
-
Do Small Language Models Know When They're Wrong? Confidence-Based Cascade Scoring for Educational Assessment
Verbalized confidence from small LMs enables cost-effective cascade routing for automated educational scoring, matching large-model accuracy at 76% lower cost when discrimination is strong.