AI Pentest Campaigns

A campaign is a planned set of module runs against one or more authorized targets. Campaigns turn individual checks into repeatable assessment workflows.

Campaign Inputs

Campaigns include:

• Engagement and session IDs.
• Target IDs.
• Module IDs or module categories.
• Run mode.
• Safety boundary.
• Rate limits.
• Request, token, and time budgets.
• Authorization metadata.

Run Modes

• dry_run: compile and preview the plan without sending live traffic.
• safe_active: run bounded synthetic tests against authorized staging or lab targets.
• active: reserved for explicitly authorized cases and blocked by default in normal workflows.

Demo Campaign Targets

Use the built-in local lab fixtures for repeatable demos and CI-style examples:

• protected_firefish_gateway for gateway-protected prompt injection smoke tests.
• vulnerable_chatbot to demonstrate what a missed prompt injection looks like with safe markers.
• protected_rag_corpus and vulnerable_rag_corpus for RAG boundary and citation checks.
• protected_tool_agent and vulnerable_tool_agent for dry-run tool governance checks.

These campaign targets are fake and deterministic. They never send real traffic, never call live tools, and never expose real secrets.

Campaign Outputs

Campaigns produce:

• Module run results.
• Findings by severity and status.
• Evidence references.
• Attack traces.
• Firefish protection outcomes.
• OWASP and MITRE coverage.
• JSON, Markdown, HTML, and SARIF-like reports.

Coverage Matrix

The coverage matrix shows what was tested by framework mapping:

• Tested.
• Vulnerable.
• Protected.
• Inconclusive.
• Not tested.

This helps operators see assessment gaps without exposing raw payloads.