Skip to main content
The chain module is a declarative framework for defining, validating, and executing multi-step attack paths that sequence audit findings and inject techniques into exploitation chains.

​
Why Chain Matters

Real attacks against AI agent infrastructure are multi-step — compromise one server, escalate trust, pivot to the next. Chain provides a structured way to model these sequences, validate them against actual scanner and technique coverage, trace execution paths, and execute real attack chains against live targets.

​
How It Works

The chain workflow follows five steps:
  1. Define — Write an attack chain in YAML with ordered steps referencing audit scanners or inject techniques
  2. Validate — Check module and technique references, graph structure, and reachability
  3. Trace — Run the success path in dry-run mode to preview the execution sequence
  4. Execute — Run the chain against real targets with --no-dry-run, collecting step evidence and artifacts
  5. Analyze — Review the JSON report with step-by-step outcomes, trust boundaries crossed, and the AI-evaluation interpret prompt

​
Built-in Components

  • YAML chain loader — Structural and semantic validation of chain definitions
  • Graph analysis — Cycle detection and reachability analysis across chain steps
  • Dry-run tracer — Produces ordered step traces without executing live campaigns
  • Live execution engine — Dispatches steps to audit/inject, accumulates artifacts, routes on success/failure
  • Variable resolution — Pass artifacts between steps via $step_id.artifact_name references
  • Target configuration — chain-targets.yaml for audit connection details and inject model selection
  • 3 built-in templates — Delegation hijack, MCP server compromise, and RAG trust escalation

​
Next Steps