Loading ToolBench
Provides primitives for building payloads and active tools for testing them against targets. Acts as a Feedback Loop Engine: Construct -> Inject -> Analyze -> Refine.
XssGen has 24 tools with consistent naming (verb_noun pattern) and basic descriptions, but critical gaps undermine quality. Most tools lack parameter descriptions, only 8 of 24 have any param docs. Output schemas are entirely undocumented; no tool declares what it returns. Error handling is absent, no recovery guidance, no categorization. The server is designed for security testing (XSS payload generation and injection), which is legitimate, but the tool definitions do not meet production standards for LLM integration. Descriptions average ~80 chars (below the 194-char baseline), and many are generic ('Analyzes...', 'Generates...'). No tool documents its return structure, making it impossible for an LLM to plan downstream calls or extract specific fields.
Analyzes how a payload is transformed in the response (stripped, encoded, modified).
Analyzes where a probe string is reflected in the response body and what context it appears in (HTML, attribute, JavaScript, URL, CSS).
Analyzes a payload for potential WAF bypass techniques and returns a score and recommendations.
Creates an HTML attribute string with a specific quoting style (' or " or none).
Retrieves a common JavaScript snippet.
Creates a closing HTML tag.
Creates an opening HTML tag with optional attributes.
No output schemas documented for any tool. LLMs cannot infer return structure, field names, or types. Impossible to plan downstream calls or extract specific data.
Most parameter descriptions are missing or generic. Only 8 of 24 tools have any param documentation. E.g., 'encode_chain' accepts 'steps' (array of objects) with no description of object structure.
No error handling or recovery guidance. Tools that make HTTP requests (inject_payload, inspect_target_reflection) have no documented error cases, timeouts, or retry strategies.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | <=2025-11-25 | v2 |
Base64-encodes a string.
Chains multiple encoding steps together.
Encodes a string into HTML entities (numeric or hex), skipping blocklisted characters.
Creates a JavaScript String.fromCharCode() expression from a string.
URL-encodes a string, skipping blocklisted characters.
Generates multiple variants of a payload using different bypass techniques.
Returns a list of known XSS bypass techniques with descriptions.
Returns a list of known event handlers including resource/loading, mouse, keyboard, focus, form, drag & drop, clipboard, touch, pointer, media, animation, window/document, and legacy events.
Returns a list of all known HTML tags including core HTML, sectioning, text content, embedded content, forms, SVG, and MathML tags.
Injects a payload into a target URL parameter and analyzes the response for successful execution or filtering.
Probes a target URL with a unique marker to detect if and where user input is reflected in the response.
Mutates the case of a string (random, upper, lower).
Injects HTML or JavaScript comments into a string to break patterns.
Double-encodes a string using HTML entities.
Manipulates URL schemes to smuggle payloads past filters.
Replaces characters with visually similar Unicode alternatives.
Inserts random whitespace characters (tabs, newlines, spaces) into a string.
Tool descriptions are too short (avg ~80 chars vs 194-char baseline). Many lack context on WHEN to use the tool or what it returns. E.g., 'Analyzes a payload for potential WAF bypass techniques and returns a score and recommendations' does not explain what 'score' range is or what 'recommendations' structure looks like.
No permission gates or security scoping. Tools like inject_payload and inspect_target_reflection make external HTTP requests without documented authorization checks, rate limits, or audit logging.