lo6

Security Specification (OWASP Compliance)

Overview

This document outlines the security measures for lo6, specifically addressing the OWASP Top 10 (2021) vulnerabilities.

OWASP Top 10 Mitigation Strategy

Vulnerability	Mitigation Strategy in Our Stack
A01: Broken Access Control	RBAC Middleware: Next.js Middleware to enforce roles on all routes. Row Level Security (RLS): Supabase RLS policies to ensure users only access data they are permitted to.
A02: Cryptographic Failures	HTTPS Everywhere: Enforced by Vercel/Deployment platform. Env Vars: Secrets managed via .env.local and never committed. Encryption: Data at rest encrypted by PostgreSQL provider.
A03: Injection	ORM Usage: Prisma/Supabase client prevents SQL injection. Zod Validation: Strict schema validation for all LLM inputs and User inputs.
A04: Insecure Design	Threat Modeling: Reviewing agent workflows for “prompt injection” risks. Rate Limiting: Upstash Redis to rate limit API routes.
A05: Security Misconfiguration	Headers: Helmet.js equivalent security headers (CSP, HSTS). Minimal Docker Image: Distroless images for container deployment.
A06: Vulnerable and Outdated Components	Dependabot: Automated dependency updates. Snyk: CI/CD pipeline scanning for vulnerable packages.
A07: Identification and Authentication Failures	Supabase Auth: Delegated secure auth (OAuth/Magic Links). MFA: Enforced for Admin/Publisher roles.
A08: Software and Data Integrity Failures	CI/CD Signing: Verified commits and build pipelines. Subresource Integrity: For any external scripts.
A09: Security Logging and Monitoring	Audit Logs: Dedicated audit_logs table for all critical actions (Publish, Delete). Alerting: Sentry for runtime errors and security exceptions.
A10: Server-Side Request Forgery (SSRF)	URL Validation: Strict allowlisting of domains. Private Network Block: Agents cannot access internal IPs. Crawler Ethics: Respect robots.txt, identify via User-Agent, and enforce per-domain rate limits.

Role-Based Access Control (RBAC)

Role	Permissions
Viewer	Read-only access to published content.
Journalist	Create Leads, Write Stories, Trigger Research Agents.
Editor	Approve Leads, Edit Stories, Reject Content.
Publisher	Publish to external platforms, Manage Configuration.
Admin	Manage Users, View Audit Logs, System Configuration.

AI-Specific Security (LLM Top 10)

• Prompt Injection: Separate system instructions from user content. Use “sandwich” defense where applicable.
• Output Handling: Treat all LLM output as untrusted. Sanitize markdown/HTML before rendering.
• Excessive Agency: Agents require “Human-in-the-Loop” approval for high-impact actions (Publishing).

Comprehensive Compliance Mapping

NIST Cybersecurity Framework (CSF)

| Function | Category | Implementation | | :— | :— | :— | | Identify | Asset Management | Automated inventory of all Agent Tools and API endpoints. | | Protect | Access Control | RBAC + MFA (Supabase). Least Privilege for Agent API keys. | | Detect | Anomalies | Real-time monitoring of Token Usage spikes (Cost/Security). | | Respond | Mitigation | Kill-switch for all Agents via “Emergency Stop” feature. | | Recover | Backups | Point-in-time recovery (PITR) enabled on PostgreSQL. |

ISO/IEC 27001:2022 Controls

• A.5.15 Access Control: Strict segregation of duties (Journalist vs Publisher).
• A.8.25 Secure Development Lifecycle: CI/CD pipeline with automated SAST (Static Application Security Testing).
• A.8.28 Secure Coding: Enforcement of SEI CERT coding standards via ESLint/SonarQube.

CIS Controls (v8) - Implementation Group 1

• Control 3 (Data Protection): All sensitive data (API Keys, PII) encrypted at rest and in transit.
• Control 4 (Secure Configuration): Infrastructure as Code (Terraform/Pulumi) to prevent drift.
• Control 11 (Data Recovery): Automated weekly backups tested for restoration.

SEI CERT C/C++ (Adapted for TypeScript)

While originally for C/C++, we adapt core principles for TypeScript:

• Input Validation: “Validate all input from untrusted sources” -> Zod Schemas for every API route.
• Error Handling: “Do not leak information in error messages” -> Generic error responses for clients, detailed logs for internal audit.
• Concurrency: “Prevent race conditions” -> Database transactions (Prisma $transaction) for multi-step agent updates.

This site is open source. Improve this page.