How shivraste23/vulnclaw works
VulnClaw is an AI-assisted penetration-testing agent with CLI, terminal-workbench, and Web UI entry points. The documented main workflow takes a natural-language objective through reconnaissance, vulnerability discovery, exploitation/verification, and report generation. The graph also preserves the product’s MCP, plugin, traffic-evidence, skills, and knowledge-base capabilities. Relationships follow README claims and sampled entry-point/tool-management code; several unsampled subsystem connections are intentionally left unspecified.
Main entry points: tool_call_manager.py Tool Dispatch · loop_controller.py Solve and Loop · orchestrator.py Run Orchestrator · findings.py Finding Analysis
Components 25
User Interfaces
- CLI and REPL
vulnclaw/cli/main.py - Terminal Workbench
tui/src/app.rs - Web UI
frontend/src/App.tsx - Web API
vulnclaw/web/app.py - Web Task Manager
vulnclaw/web/task_manager.py
Agent Runtime
- Run Orchestrator
vulnclaw/orchestrator.py - Autonomous Agent
vulnclaw/agent/core.py - Solve and Loop
vulnclaw/agent/loop_controller.py - LLM Client
vulnclaw/agent/llm_client.py - Tool Dispatch
vulnclaw/agent/tool_call_manager.py
Security Capabilities
- Reconnaissance
vulnclaw/agent/recon_tools.py - Vulnerability Intelligence
vulnclaw/intel/tools.py - Detection Plugins
vulnclaw/plugins/runtime.py - Skills and Knowledge
vulnclaw/skills/dispatcher.py
Evidence and Reports
- Traffic Evidence
vulnclaw/traffic/store.py - Finding Analysis
vulnclaw/intel/findings.py - Report and PoC
vulnclaw/report/generator.py - Report Export
vulnclaw/report/pdf_exporter.py
Integrations and State
- MCP Services
vulnclaw/mcp/router.py - External MCP Tools
- Model Provider
- Run and Target State
vulnclaw/target_state/store.py - Knowledge Base
vulnclaw/kb/retriever.py
Other components
- Operator
- Authorized Target
Connections 30
- Operator → CLI and REPL: submits objective
- Operator → Terminal Workbench: submits objective
- Operator → Web UI: submits objective
- Web UI → Web API: calls API
- Web API → Web Task Manager: starts task
- CLI and REPL → Run Orchestrator: starts run
- Web Task Manager → Run Orchestrator: runs task
- Terminal Workbench → Run Orchestrator: executes command
- Run Orchestrator → Autonomous Agent: creates agent
- Autonomous Agent → Solve and Loop: solves objective
- Solve and Loop → LLM Client: requests reasoning
- LLM Client → Model Provider: sends completion request
- Solve and Loop → Tool Dispatch: dispatches tool calls
- Tool Dispatch → MCP Services: executes MCP tool
- MCP Services → External MCP Tools: dispatches requests
- Tool Dispatch → Reconnaissance: invokes tools
- Tool Dispatch → Vulnerability Intelligence: invokes tools
- Tool Dispatch → Detection Plugins: runs plugins
- Tool Dispatch → Skills and Knowledge: loads references
- Tool Dispatch → Traffic Evidence: records traffic evidence
- Traffic Evidence → Tool Dispatch: serves evidence queries
- Tool Dispatch → Run and Target State: records run state
- Solve and Loop → Finding Analysis: parses findings
- Finding Analysis → Report and PoC: supplies findings
- Run Orchestrator → Report and PoC: generates reports
- Report and PoC → Report Export: exports PDF
- Solve and Loop → Knowledge Base: retrieves context
- Reconnaissance → Authorized Target: probes target
- Vulnerability Intelligence → Authorized Target: analyzes target evidence
- Detection Plugins → Finding Analysis: contributes results