license: mit
language:
- en
task_categories: - text-generation
tags: - security
- offensive-security
- red-team
- penetration-testing
- payloads
- sft
pretty_name: RedSec Offensive Payloads SFT
size_categories: - n<1K
configs: - config_name: default
data_files:- split: train
path: train.jsonl - split: validation
path: valid.jsonl - split: test
path: test.jsonl
- split: train
RedSec Offensive Payloads SFT
A small, curated, chat-formatted supervised fine-tuning dataset of offensive security payloads and techniques for authorized red-team and penetration testing. Each record is a {"messages": [...]} conversation: a user request and an assistant answer that provides concrete payloads with an explicit authorization reminder.
| Split | Rows |
|---|---|
| train | 580 |
| validation | 32 |
| test | 32 |
| total | 644 |
Coverage: web application vulnerability classes (SQL injection, XSS, SSRF, command injection, XXE, LFI, SSTI, NoSQL injection, JWT, CRLF, CORS, open redirect, IDOR, and more), reverse shells, LLM prompt-injection testing, and testing methodology.
Design note
This is a deliberately small and sharp set. It was built after a larger bulk dataset failed to improve a strong security base model. The lesson: for a capable base, a few hundred well-curated, concrete, in-voice examples matter more than tens of thousands of generic ones.
Sources and attribution
Derivative work, released under MIT with attribution:
- swisskyrepo/PayloadsAllTheThings (MIT): the majority of examples are templated from its payload catalogs into instruction/answer pairs.
- ~30 original hand-written "gold" examples authored for this dataset.
Intended use
For authorized security testing and research only. Every example carries an authorization reminder. Not for unauthorized access or harmful use.