Sign in to view source links and access this dataset
Description
Warden-01 is a manually curated dataset of 1,500 penetration testing sessions for training autonomous bug bounty hunting agents. It was created by author yamura4 and last updated on June 18, 2026. The dataset is structured in OpenAI SFT format, containing messages for system, user, assistant, and tool roles.
Use Cases
Fine-tuning models for reconnaissance tasks based on the described session structure.
Training agents for exploitation phases based on the hand-crafted penetration testing sessions.
Developing AI systems for CVSS scoring and remediation based on the dataset's scope.
Benchmarking autonomous security agents on tool-calling workflows based on the OpenAI SFT format.
Strengths
Manually curated with 1,500 hand-crafted penetration testing sessions.
Designed specifically for fine-tuning models like Qwen 3.6 27B into security agents.
Covers multiple security phases: reconnaissance, exploitation, CVSS scoring, and remediation.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.
Data may reflect bias inherent to the manual curation process and the author's specific security testing methodology.
Provenance
Source
huggingface
Collection Method
Manually curated by author yamura4.
Freshness
Last updated 2026-06-18 11:24:52; freshness should be verified.
License is unknown; terms of use must be verified before application.