Security Audits: Agent Traces for Open-Source Software

A collection of agent traces generated with Swival, an agent designed for open-source models. The traces focus on security audits of open-source software. The dataset was uploaded by author jedisct1 and was last updated on June 8, 2026.

Use Cases

• Analyzing agent reasoning patterns for security vulnerability detection based on the described conversation traces.
• Training or fine-tuning models for automated code review based on the security audit traces.
• Benchmarking the performance of different AI agents on security-focused software engineering tasks.
• Studying the interaction patterns between an AI agent and a codebase during an audit session.

Strengths

• Traces are generated by a specific agent system (Swival) designed for open-source models, providing a consistent source.
• Focus is explicitly on security audits of open-source software, a defined application domain.
• Data was last updated on June 8, 2026, indicating recent activity.

Limitations

• The number of traces, rows, and overall dataset size is unknown, limiting suitability assessment.
• Column-level documentation is absent; the structure and semantics of the trace data must be inferred after download.
• The description references a full description on an external page, requiring a click-through for complete context.

Provenance

Source

huggingface

Collection Method

Exported conversation traces from the Swival agent system.

Freshness

Last updated 2026-06-08 11:59:16