Data Flow

This page describes how data moves through the extension, from user action to AI response.

Chat Flow (Context Menu → AI Response)

User right-clicks request        Context menu action triggered
        │                                    │
        ▼                                    ▼
┌───────────────┐              ┌──────────────────────┐
│ Burp Selection │──────────►  │  ContextCollector     │
│ (request/issue)│              │  Extracts URL, method,│
└───────────────┘              │  headers, params, body│
                                └──────────┬───────────┘
                                           │
                                           ▼
                                ┌──────────────────────┐
                                │  Redaction Pipeline   │
                                │  Applies privacy mode │
                                │  (STRICT/BALANCED/OFF)│
                                └──────────┬───────────┘
                                           │
                                           ▼
                                ┌──────────────────────┐
                                │  Prompt Bundle        │
                                │  Context + template + │
                                │  SHA-256 hash         │
                                └──────────┬───────────┘
                                           │
                                           ▼
                                ┌──────────────────────┐
                                │  Backend Adapter      │
                                │  CLI (subprocess) or  │
                                │  HTTP (API call)      │
                                └──────────┬───────────┘
                                           │
                                    ┌──────┴──────┐
                                    ▼             ▼
                            ┌────────────┐ ┌────────────┐
                            │ Audit Log  │ │ Chat Panel │
                            │ (JSONL)    │ │ (Markdown) │
                            └────────────┘ └────────────┘

Step-by-Step

1. User selects context: Right-click on a request/response in Proxy, Repeater, or Site Map; or on an issue in the Scanner panel.

2. Context collection: ContextCollector captures the selected request/response or issue data as raw text plus metadata (URL, method, severity, confidence, remediation).

3. Redaction: The Redaction module applies the active privacy mode:

   • STRICT: Strips cookies, redacts auth tokens, anonymizes hostnames using SHA-256(salt + host).

   • BALANCED: Strips cookies, redacts auth tokens, preserves hostnames.

   • OFF: No modification.

4. Prompt bundle: The redacted context is combined with the selected prompt template. If determinism is enabled, context ordering is stabilized. A SHA-256 hash of the bundle is computed for audit integrity.

5. Backend processing: The prompt is sent to the configured AI backend (CLI subprocess or HTTP API). The response streams back in real time.

6. Audit logging: If enabled, the prompt bundle, context hashes, backend metadata, and streamed response chunks are written to ~/.burp-ai-agent/audit.jsonl.

7. Display: The AI response is rendered as Markdown in the chat panel. For scanner findings, issues may be created automatically.

Passive Scanner Flow

Proxy traffic ──► MIME filter ──► Scope filter ──► Size filter ──► Rate limiter ──► AI Analysis ──► Findings
                                                                                                     │
                                                                                        (confidence ≥ 85%)
                                                                                                     │
                                                                                                     ▼
                                                                                         Burp Issues [AI Passive]
                                                                                                     │
                                                                                          (if auto-queue enabled)
                                                                                                     │
                                                                                                     ▼
                                                                                            Active Scanner Queue

Active Scanner Flow

Target request ──► Injection Point Extraction ──► Payload Selection ──► Send Payloads ──► Response Analysis
                        │                              │                                        │
                   URL params               Risk level + scan mode                     Detection method
                   Body params              Vuln class filter                          (error/blind/time/
                   Headers                                                              reflection/OOB)
                   Cookies                                                                      │
                   JSON fields                                                                  ▼
                   XML elements                                                        Confirmed findings
                   Path segments                                                       with [AI Active] prefix

MCP Tool Flow

External AI Agent ──► SSE/STDIO ──► Token Auth (external only) ──► Tool Gating ──► Request Limiter ──► Tool Handler
        │                                                                                    │
   (Claude Desktop,                                                                   Burp API call
    custom scripts)                                                                  (history, repeater,
                                                                                      scanner, etc.)
                                                                                          │
                                                                                          ▼
                                                                                   Privacy filter
                                                                                          │
                                                                                          ▼
                                                                                   Response to agent