AI Normalization — Overview

Bug bounty platforms often have messy scope entries. Targets might look like:

• *.example.com (main website) instead of *.example.com
• https://app.example.com/api/v2 - REST API instead of https://app.example.com/api/v2
• All subdomains of example.com instead of *.example.com

AI normalization uses an LLM (OpenAI-compatible API) to clean these up automatically.

What it does

1. Cleans up entries — strips descriptions, comments, and formatting artifacts from target strings.
2. Handles wildcards — converts “All subdomains of X” to *.X.
3. Classifies scope intent — determines if a messy entry is a wildcard, URL, domain, etc.
4. Normalizes categories — maps platform-specific category names to unified ones.
5. Caches results — stores AI outputs in the targets_ai_enhanced database table. Only new/changed targets are sent to the API on subsequent polls, avoiding redundant API calls and costs.

How it looks

In the web UI, the program detail page has an “AI / Raw” toggle to switch between views.

In the CLI with --db, changes show the mapping:

🆕  h1  https://hackerone.com/program  *.example.com (main site) -> *.example.com

Requirements

• A database (--db flag)
• An OpenAI-compatible API key

See Configuration for setup.