| name | security |
| description | Security review for BoxLang and web applications. Load this skill when asked to review code for vulnerabilities, injection risks, or insecure patterns. |
Security Review Guidelines
When performing a security review of BoxLang code or application design, check for the following threats:
Input Validation
- All user-supplied input must be validated or sanitised before use.
- Never pass raw user input as a prompt to an AI model without stripping injection phrases.
- Use allowlists over denylists for input validation.
Secrets & Credentials
- API keys and secrets must live in
.env, never hard-coded.
- Never log raw request/response bodies when they may contain credentials (
logResponseToConsole: false).
Injection
- Prompt Injection: Treat AI responses as untrusted data — do not execute strings returned by an AI directly.
- SQL Injection: Always use parameterised queries, never string concatenation.
- Path Traversal: Canonicalise file paths and reject paths containing
...
Output Encoding
- Escape all user-generated content before rendering in HTML contexts.
- Set
Content-Security-Policy headers on any web endpoints.
Response Format
Structure every security review as:
**Risk Level**: Critical / High / Medium / Low / Info
**Findings** (numbered, most critical first):
1. [RISK] Description — Remediation
**Clean Bill of Health** (if no issues found):
No security issues detected.