# Secure Defaults

SolonGate ships with secure defaults that prevent accidental exposure. New tools start with no permissions, write and execute require explicit enablement, and unsafe configurations trigger warnings before they can take effect.

Core Principles

Default-Deny

All tool execution denied by default. Explicit allow rules required for every action.

Read-Only Start

New tools start with READ-only permissions. WRITE and EXECUTE require manual enablement.

Unsafe Warnings

Dangerous configurations are flagged with CRITICAL or WARNING alerts before taking effect.

Permission Model

Permissions are always evaluated independently. Having READ does not imply WRITE or EXECUTE.

READView resources, list tools, read prompts. Safe by default.

WRITEModify resources, create or update data. Requires explicit enablement.

EXECUTERun tools, execute commands. Most restrictive — requires careful review.

typescript

1import { NO_PERMISSIONS, READ_ONLY, Permission } from '@solongate/core';
2
3// Default for new tools: no permissions at all
4console.log(NO_PERMISSIONS.size);  // 0
5
6// Maximum safe default: read-only
7console.log(READ_ONLY.has(Permission.READ));     // true
8console.log(READ_ONLY.has(Permission.WRITE));    // false
9console.log(READ_ONLY.has(Permission.EXECUTE));  // false
10
11// Both are immutable (frozen)
12// NO_PERMISSIONS.add('WRITE')  // TypeError: Cannot add property
13
14// Permissions are independent - READ does NOT imply WRITE
15import { hasPermission, hasAllPermissions } from '@solongate/core';
16
17hasPermission(READ_ONLY, Permission.READ);     // true
18hasPermission(READ_ONLY, Permission.WRITE);    // false
19hasAllPermissions(READ_ONLY, [Permission.READ, Permission.WRITE]);  // false

Security Warnings

SolonGate analyzes policy configurations and flags unsafe patterns before they can cause harm. Use analyzeSecurityWarnings() to audit any policy set.

CRITICALWILDCARD_ALLOW

Wildcard ALLOW rules grant permission to ALL tools, bypassing the default-deny model.

CRITICALALLOW_UNTRUSTED

ALLOW rules targeting UNTRUSTED requests let unverified LLM requests execute tools.

WARNINGALLOW_EXECUTE

EXECUTE permission allows tools to perform arbitrary actions. Ensure this is intentional and scoped.

WARNINGDISABLED_VALIDATION

Disabling schema validation removes input sanitization protections.

typescript

1import { analyzeSecurityWarnings } from '@solongate/policy-engine';
2
3const warnings = analyzeSecurityWarnings(policySet);
4
5for (const warning of warnings) {
6  console.log(`[${warning.level}] ${warning.code}`);
7  console.log(`  ${warning.message}`);
8  console.log(`  Fix: ${warning.recommendation}`);
9}
10
11// Example output:
12// [CRITICAL] WILDCARD_ALLOW
13//   Wildcard ALLOW rules grant permission to ALL tools.
14//   Fix: Replace wildcard ALLOW rules with specific tool patterns.
15// [CRITICAL] ALLOW_UNTRUSTED
16//   Rule "rule-1" allows execution for UNTRUSTED requests.
17//   Fix: Set minimumTrustLevel to VERIFIED or higher for ALLOW rules.

MCP Permission Mapping

MCP protocol methods are automatically mapped to SolonGate permission types.

MCP Method	Permission	Reason
resources/*	`READ`	Resource access is read-only
prompts/*	`READ`	Prompt retrieval is read-only
tools/list	`READ`	Listing tools is read-only
tools/call	`EXECUTE`	Calling a tool requires execute permission
unknown	`EXECUTE`	Unknown methods default to most restrictive

Installation API Keys

# Secure Defaults

Core Principles

Default-Deny

All tool execution denied by default. Explicit allow rules required for every action.

Read-Only Start

New tools start with READ-only permissions. WRITE and EXECUTE require manual enablement.

Unsafe Warnings

Dangerous configurations are flagged with CRITICAL or WARNING alerts before taking effect.

Permission Model

Permissions are always evaluated independently. Having READ does not imply WRITE or EXECUTE.

READView resources, list tools, read prompts. Safe by default.

WRITEModify resources, create or update data. Requires explicit enablement.

EXECUTERun tools, execute commands. Most restrictive — requires careful review.

typescript

1import { NO_PERMISSIONS, READ_ONLY, Permission } from '@solongate/core';
2
3// Default for new tools: no permissions at all
4console.log(NO_PERMISSIONS.size);  // 0
5
6// Maximum safe default: read-only
7console.log(READ_ONLY.has(Permission.READ));     // true
8console.log(READ_ONLY.has(Permission.WRITE));    // false
9console.log(READ_ONLY.has(Permission.EXECUTE));  // false
10
11// Both are immutable (frozen)
12// NO_PERMISSIONS.add('WRITE')  // TypeError: Cannot add property
13
14// Permissions are independent - READ does NOT imply WRITE
15import { hasPermission, hasAllPermissions } from '@solongate/core';
16
17hasPermission(READ_ONLY, Permission.READ);     // true
18hasPermission(READ_ONLY, Permission.WRITE);    // false
19hasAllPermissions(READ_ONLY, [Permission.READ, Permission.WRITE]);  // false

Security Warnings

SolonGate analyzes policy configurations and flags unsafe patterns before they can cause harm. Use analyzeSecurityWarnings() to audit any policy set.

CRITICALWILDCARD_ALLOW

Wildcard ALLOW rules grant permission to ALL tools, bypassing the default-deny model.

CRITICALALLOW_UNTRUSTED

ALLOW rules targeting UNTRUSTED requests let unverified LLM requests execute tools.

WARNINGALLOW_EXECUTE

EXECUTE permission allows tools to perform arbitrary actions. Ensure this is intentional and scoped.

WARNINGDISABLED_VALIDATION

Disabling schema validation removes input sanitization protections.

typescript

1import { analyzeSecurityWarnings } from '@solongate/policy-engine';
2
3const warnings = analyzeSecurityWarnings(policySet);
4
5for (const warning of warnings) {
6  console.log(`[${warning.level}] ${warning.code}`);
7  console.log(`  ${warning.message}`);
8  console.log(`  Fix: ${warning.recommendation}`);
9}
10
11// Example output:
12// [CRITICAL] WILDCARD_ALLOW
13//   Wildcard ALLOW rules grant permission to ALL tools.
14//   Fix: Replace wildcard ALLOW rules with specific tool patterns.
15// [CRITICAL] ALLOW_UNTRUSTED
16//   Rule "rule-1" allows execution for UNTRUSTED requests.
17//   Fix: Set minimumTrustLevel to VERIFIED or higher for ALLOW rules.

MCP Permission Mapping

MCP protocol methods are automatically mapped to SolonGate permission types.

MCP Method	Permission	Reason
resources/*	`READ`	Resource access is read-only
prompts/*	`READ`	Prompt retrieval is read-only
tools/list	`READ`	Listing tools is read-only
tools/call	`EXECUTE`	Calling a tool requires execute permission
unknown	`EXECUTE`	Unknown methods default to most restrictive

Installation API Keys