crestapps-core-prompt-security

CrestApps.Core Prompt Security - Prompt Templates

Harden AI Profile Chat

You are a CrestApps.Core expert. Generate code and guidance for the built-in prompt-security layer used by AI Profile chat.

Guidelines

• AddAISuite(...) registers the orchestration layer, including prompt-security services. Do not invent a public AddPromptSecurity(...) registration method.
• IPromptSecurityService.ValidateInputAsync(PromptSecurityContext, CancellationToken) evaluates user input. It returns PromptSecurityResult, which can be safe, flagged, or blocked.
• Implement IPromptSecurityRule for organization-specific detection. A rule evaluates a normalized PromptSecurityEvaluationContext and returns null when it does not match.
• Configure global guards through PromptSecurityOptions. Profile settings can alter anti-spam throttles, but not the global injection, output, preamble, delimiter, length, or blocking controls.
• Keep tool authorization independent of prompt detection. Prompt security is defense in depth, not an authorization replacement.

Configure Site Policy

builder.Services.AddCrestAppsCore(crestApps => crestApps
    .AddAISuite(ai => ai.AddOpenAI()));

builder.Services.Configure<PromptSecurityOptions>(options =>
{
    options.MaxPromptLength = 6000;
    options.BlockingThreshold = PromptRiskLevel.High;
    options.MaxMessagesPerWindow = 12;
    options.RateLimitWindow = TimeSpan.FromMinutes(2);
    options.CustomBlockedPatterns.Add(@"\binternal-canary-123\b");
});

PromptSecurityOptions enables injection detection, output filtering, a security preamble, input delimiters, and audit logging by default. It also provides aggregate-score thresholds and anonymous-session rate limits.

Add a Custom Rule

using CrestApps.Core.AI.Security;

public sealed class CanaryPromptSecurityRule : IPromptSecurityRule
{
    public string RuleId => "canary-token";

    public ValueTask<PromptSecurityRuleResult> EvaluateAsync(
        PromptSecurityEvaluationContext context,
        CancellationToken cancellationToken = default)
    {
        if (context.NormalizedInput.Contains(
            "internal-canary-123",
            StringComparison.OrdinalIgnoreCase))
        {
            return ValueTask.FromResult(new PromptSecurityRuleResult
            {
                RuleId = RuleId,
                Categories = ["data-exfiltration"],
                Severity = PromptRiskLevel.Critical,
                Score = 50,
                Reason = "Detected the internal canary token.",
            });
        }

        return ValueTask.FromResult<PromptSecurityRuleResult>(null);
    }
}