Skip to main content
Enclave provides pre-configured security profiles that balance functionality against risk. Choose the appropriate level based on your trust model and use case.

Using Security Levels

Security Level Comparison

The level also selects the globals a script may use (STRICT has no parseInt, parseFloat, isNaN, isFinite and URI functions; console without a consoleHandler exists only at PERMISSIVE), what the runtime refuses on every value (secureProxyConfig), and the defaults of the worker pool, the reference sidecar and the Babel transform. memoryLimit is 1 MB at every level. sanitizeStackTraces concerns errors made outside the sandbox: an error made inside it has no stack at any level (see Configuration).
The exported SECURITY_LEVEL_CONFIGS table also has the fields blockTimingAPIs, allowUnboundedLoops and unicodeSecurityCheck. The enclave does not act on them: Date is available at every level, the agentscript preset refuses while and do...while loops at every level, and the Unicode checks are those of the validator’s pre-scanner, which do not depend on the level.

When to Use Each Level

STRICT

Use for maximum security with untrusted code:
  • AI-generated scripts from external sources
  • User-submitted code
  • Third-party plugin code
  • Any code where you cannot verify the source

SECURE

Balanced security for semi-trusted scenarios:
  • Internal automation scripts
  • Validated AI-generated code
  • Scripts from authenticated users

STANDARD

Default level for trusted internal use:
  • Internal workflow automation
  • Development and testing
  • Scripts you control

PERMISSIVE

Minimal restrictions for controlled environments:
  • Internal testing
  • Performance benchmarking
  • Trusted scripts where you need maximum flexibility
Use STRICT for any untrusted code, including AI-generated scripts from external sources or user-submitted scripts.

Customizing Security Levels

You can start with a preset and override specific settings:

Defense-in-Depth

Security levels configure multiple defense layers:
  1. AST Validation - The allowed globals vary by level
  2. Resource Limits - Timeout, iterations, tool calls
  3. Output Controls - Console rate limiting, output size, the depth and size of sanitized values
  4. Stack Sanitization - For the errors that carry a stack
Each level provides appropriate defaults for its trust model.