Agent System

The multi-agent orchestration system distributes offensive operations across 15 specialized AI sub-agents. Each agent owns a specific domain and operates with focused instructions and tool access.

Agent Categories

Coordinators (3)

Agents that plan, delegate, and correlate findings.

Agent	Role
Pentest Orchestrator	Full-engagement coordination, agent delegation
Attack Path Architect	Kill chain design, MITRE ATT&CK mapping
Vuln Chain Composer	Multi-step exploit chain correlation

Specialized Executors (11)

Agents that execute specific attack types.

Agent	Domain
SQL Injection Executor	SQLi detection and exploitation
XSS Executor	Cross-site scripting
SSRF Executor	Server-side request forgery
Auth Bypass Executor	Authentication bypass, JWT, OAuth
API Security Executor	REST, GraphQL, WebSocket testing
Client-Side Executor	CSPT, DOM injection, postMessage
Server-Side Executor	SSTI, deserialization, command injection
Cloud Attack Executor	S3, IAM, metadata, serverless
Post-Exploitation Executor	Privesc, lateral movement, persistence
Crypto Audit Executor	Smart contracts, DeFi, bridge security
Blockchain Exploit Executor	Chain-specific vulnerability exploitation

Support (3)

Agents that provide infrastructure and utility.

Agent	Role
Recon Dominator	Subdomain, port, tech fingerprinting
HackerOne Hunter	Bug bounty workflow automation
Test Detection	Target quality filtering, honeypot detection

Execution Workflow

4-Phase Executor Loop

RECON --> Identify attack surface, enumerate endpoints, fingerprint technologies
   |
EXPERIMENT --> Generate payloads and techniques tailored to findings
   |
TEST --> Apply payloads against live targets, record responses
   |
VERIFY --> Confirm findings, eliminate false positives, produce PoC

5-Phase Post-Exploitation Loop

STABILIZE --> Establish reliable access, deploy persistence
   |
ESCALATE --> Privilege escalation (Linux/Windows/AD)
   |
MOVE --> Lateral movement, pivot through networks
   |
PERSIST --> Deploy implants, backdoors, scheduled tasks
   |
EXFILTRATE --> Collect, classify, compress, and exfiltrate loot

Delegation Rules

• Single target, single vuln type --- deploy specialized executor directly. No orchestrator overhead.
• Single target, unknown vulns --- single orchestrator spawns multiple executors.
• Multiple targets --- orchestrator spawns one executor per target.
• Bug bounty --- HackerOne Hunter spawns orchestrators per asset group.
• Maximum nesting depth: 2. Orchestrators do not spawn orchestrators beyond one level.

Execution Modes

Mode	Description
Foreground	Inline execution with real-time output
Background	Detached execution with notification on completion
Worktree isolated	Git worktree sandbox for parallel, non-interfering execution

Communication

Agents communicate through the engagement state in the Knowledge Base. Findings flow from executors to the pipeline. The orchestrator reads accumulated findings and dispatches follow-on agents. No agent blocks waiting for another -- the engagement state provides eventual consistency.