Frankestain

Private instruction-tuning dataset for a personal cybersecurity study assistant aligned to a phased learning roadmap (Linux/networking → scripting → low-level → structured paths). Authorized lab use only. Content is for learning on systems you own or have explicit permission to test. Do not use it to attack networks, devices, or accounts you do not own.

Dataset summary

• Format: JSON array of objects
• Fields:
  • instruction (string) — user/ask prompt
  • output (string) — target assistant reply
  • topic (string) — content bucket (e.g. bandit_system_skills, linux_permissions_and_files, duckyscript_advanced)
  • phase (string) — roadmap phase 1–4
  • side (string) — blue or red framing
• Language mix: Italian and English prompts/answers appear in the data
• Scale: hundreds of examples (small / growing); not a production-scale corpus

Phases (roadmap alignment)

Phase	Focus (high level)
1	Linux basics: shell, files, permissions, processes, Bandit-style skills
2	Networking / scripting-oriented security topics
3	Lower-level and specialized topics (e.g. analysis, HID/DuckyScript lab material)
4	Broader structured-path style material
Topic labels should match the phase. Phase 1 is reserved for foundational Linux skills, not advanced Windows exploitation.

Intended use

• Supervised fine-tuning (SFT) / LoRA of a small assistant that explains concepts and lab workflows
• Filtering by phase, topic, or side for curriculum-specific adapters
• Holding out an evaluation split (do not train on every row)

Out of scope

• Not a substitute for hands-on practice (terminal, labs, hardware you own)
• Not a source of “copy-paste attack” playbooks for unauthorized use
• Not sufficient alone for strong code generation (e.g. C) or expert DuckyScript without many more dedicated examples and hardware validation

Data quality notes

• Prefer diverse, non-duplicated examples over near-copies
• Keep phase / topic consistent with difficulty
• Document lab-only assumptions in answers where actions are described

File layout

• merged_dataset_all.json — main training file
• README.md — this card

Citation / ownership

Private dataset maintained by the repo owner for personal education. Redistribution only as allowed by the owner.