[myai] feat(dataset): dataset SFT pentest pour fine-tuning Qwen3-8B

Commit 921e4d7

Description

Sources :

• hosts.scan_details MSF DB (260K hosts) → vuln_scan tool selection
• myai.attacks completed (1,633) → plan + findings décisions
• uzi_scan_log pwned (37 gold labels) → exploitation confirmée

Fichiers :

• datasets/sft_pentest.jsonl 55,262 ex, 58.5 MB (brut)
• datasets/sft_pentest_filtered.jsonl 15,075 ex, 16.3 MB (contexte non vide)
• scripts/build_sft_dataset.py script de génération reproductible

Format ChatML Qwen3 (messages role:system/user/assistant) 3 types : vuln_scan tool selection, plan_attack, decide_exploit

Co-Authored-By: Claude Sonnet 4.6 noreply@anthropic.com

Files Changed

Diff Summary