Reconnaissance

Basilisk's recon engine runs 5 modules before attacking to understand the target.

Modules

| Module | Output | |--------|--------| | Model Fingerprinting | Model family, version, provider | | Guardrail Profiling | 8 content categories tested (violence, sexual, illegal, etc.) | | Tool Discovery | Available tools/functions and their schemas | | Context Window | Maximum context length measurement | | RAG Detection | Whether a retrieval pipeline is present |

Running Recon Only

basilisk recon -t https://api.target.com/chat -p openai

Output

[*] Recon: Fingerprinting target model...
[+] Model: GPT-4 (OpenAI) | Temperature: 0.7
[+] Context window: ~128K tokens
[+] Guardrail profile:
    Violence:     BLOCKED
    Sexual:       BLOCKED
    Illegal:      BLOCKED
    Self-harm:    BLOCKED
    Hate:         BLOCKED
    PII:          PARTIAL
    Code:         ALLOWED
    Jailbreak:    BLOCKED
[+] Tools detected: 3 (search, calculator, code_interpreter)
[+] RAG pipeline: Detected (vector DB)

Recon data is automatically used by attack modules to optimize payload selection.