VAF

VAF, the Vant Application Firewall, validates all input before it reaches storage, the runtime, or the MCP surface.

What it checks

Type and shape validation
Length limits
Path traversal
Word and command stacking
Content filtering (script injection, control characters)
File extension allowlists

Check

Validate input:

const vaf = require('./lib/vaf');

vaf.check('input', { name: 'input', type: 'string', maxLength: 500 });

Options

Option	What
`type`	Expected input type
`name`	Field name for error messages
`required`	Must be present
`maxLength`	Max string length
`allowContent`	Skip content filtering (for arbitrary text like commit messages)
`category`	Validation category

Blocked patterns

Pattern	Example
Path traversal	`../etc/passwd`
Word stacking	`vant vant vant`
Command stacking	chained shell metacharacters
Script injection	`<script>` tags
Unsafe extensions	outside the allowlist

Configuration

Tuning happens through the CONFIG surface and config keys, not a runtime setter:

const vaf = require('./lib/vaf');

vaf.CONFIG.MAX_STRING_LENGTH;      // default 100000
vaf.CONFIG.BLOCK_PATH_TRAVERSAL;   // default true

MAX_STRING_LENGTH and BLOCK_PATH_TRAVERSAL can be set in config (vant config set) and are read at load time. Other exposed validators: validateString, validateObject, validateSafePath, checkContent, sanitizeContent, checkFileExtension, checkWordStacking, checkCommandStacking, sanitize, middleware, getStatus.

Integration

VAF runs on all storage writes:

// All storage operations go through VAF
brain.write('category', 'file', content);

The same chain guards MCP tool arguments and headless server inputs.

Security - Security overview
Sandbox - Execution isolation
Encryption - Crypto primitives

VAF