en un clic
en un clic
Use WebQA to test websites, web pages, URLs, login flows, search flows, forms, navigation, and core user journeys with an AI browser QA agent.
Traverse all interactive elements on the page — click clickables,
Decompose a task into steps with completion checkpoints.
Structured error recovery for failed or ineffective browser
UI audit for hierarchy, accessibility, and UX.
| name | nuclei-scan |
| description | Run a Nuclei security scan against the target URL and report |
Run a security vulnerability scan against the target URL using the built-in execute_nuclei_scan tool, and produce a structured security report.
Important: Before each tool call, output a short one-line description of what you are about to do (e.g. "Running smoke security scan on target URL"). This helps generate readable step-by-step reports.
Task wording: If the user asks for 基础 / 初步 / 基础安全漏洞 / 快速 / 冒烟 / CI 门禁-style coverage, assume they want time over breadth and use mode="smoke" (and state in the report that coverage is intentionally reduced). Only use mode="deep" when they ask for 全面 / 深度 / 穷尽 scans or comparable wording.
execute_nuclei_scan tool must be available (check the tool list).Call execute_nuclei_scan with the target URL. You do not need to write raw shell commands; the tool automatically handles nuclei subprocess execution, JSONL parsing, and result summarization.
Choose a smoke or deep mode based on the user's prompt.
Example:
{
"url": "https://example.com",
"scan_types": "xss,sqli,cve",
"mode": "smoke"
}
Parameter notes:
url — the exact target URL.scan_types — Comma-separated classes of vulnerabilities (cve,xss,sqli,misconfig,exposure).mode — "smoke" for a fast scan (omits OAST templates, shorter timeouts), "deep" for a comprehensive scan (slower).Do not run a second full Nuclei scan against the same URL and the same (or broader) -tags just to “confirm how many hits there are” or to “see if there was only one finding.” The first scan's output is the source of truth.
The execute_nuclei_scan tool returns a pre-formatted, translated text summary grouped by severity (Critical, High, Medium, Low, Info), including template IDs, names, and affected URLs.
Structure your final step response to the user by echoing this formatted text. For example:
安全扫描完成(Nuclei):共发现 N 个问题。
● Critical (1个):
- Log4j RCE (CVE-2021-44228) — https://example.com/api/login
● High (2个):
- SQL Injection (sqli-detect) — https://example.com/search?q=
- Reflected XSS (xss-detect) — https://example.com/error?msg=
● Medium (1个):
- Missing X-Frame-Options header (x-frame-options) — https://example.com/
扫描模式:smoke
扫描范围:tags=xss,sqli,cve
If no findings:
安全扫描完成(Nuclei):未发现已知漏洞。
扫描模式:smoke
扫描范围:tags=xss,sqli,cve
Set your overall test status to:
failed if any Critical or High findings.warning if only Medium/Low findings.passed if no findings.[FTL] no templates provided for scanIf the tool returns a warning that nuclei failed because there were no templates provided, it means the running environment (Docker container or local host) has an empty template library.
There is no in-agent fix for this; the environment administrator must ensure nuclei -update-templates succeeds during image build or initialization. In this case, report the failure as a [warning] and explain that the environment lacks nuclei templates.