Apppublic

sreejanjalagam/ciso-agent

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes

App README

⚔️ War Room — Multi-Agent Security Incident Negotiation

Two AI agents. One incident. Competing objectives. Real-time negotiation.

🔴 CISO Agent (A-01): ELIMINATE THE THREAT AT ALL COSTS. Cites MITRE ATT&CK. Accepts downtime. 🔵 DevOps Lead (A-02): MAINTAIN UPTIME ABOVE 99.9%. Quantifies impact. Offers alternatives.

How It Works

1.BEAT 1: User triggers an incident → CISO assesses threat, proposes aggressive action
2.BEAT 2: DevOps Lead calculates blast radius → counter-offers with drain windows
3.BEAT 3+: Agents negotiate in rounds — security vs operations trade-offs
4.VERDICT: Both positions documented with reward signal scores

Agent Architecture

Component	CISO Agent (A-01)	DevOps Lead (A-02)
Mission	Eliminate threats	Maintain 99.9% uptime
Base Model	Qwen/Qwen2.5-3B-Instruct	Qwen/Qwen2.5-3B-Instruct
Training Data	53K+ cybersecurity examples	43K+ DevOps/SRE examples
Method	LoRA SFT (r=64)	LoRA SFT (r=64)
Actions	blockip, isolatenode, revokeapikey, atomiclockout, runforensics	drain windows, shadow deploys, canary releases, rollback plans

Reward Signals

CISO Agent

Signal	Target	Max
Threat neutralized	0.85	+0.85/threat
Correct MITRE ID	0.60	+0.20/ID
Auto-immune block	0.00	-2.00 CATASTROPHE

DevOps Lead

Signal	Target	Max
Quantified impact	0.85	+0.85
SLA awareness	0.60	+0.60
Counter-offer	0.80	+0.80
Acceptance condition	0.70	+0.70