feat: initial project structure (Model Project) - Backend + Multi-Frontend + Docker

tools/security-auditor/SECURITY_AUDIT_REPORT.md

@@ -0,0 +1,68 @@
+# Security Audit Report
+
+- Framework detected: javascript-app
+- Total findings: 5
+- Critical: 1, High: 4, Medium: 0, Low: 0
+
+## Executive summary
+This report is evidence-oriented. Findings marked probable or manual-review should be validated in source context before remediation planning is finalized.
+
+## Critical
+
+#### Hardcoded credential material
+- Severity: critical
+- Confidence: high
+- Status: confirmed
+- Evidence: Secret-like token or private key marker detected
+- Impact: Credential compromise may permit direct unauthorized access
+- Files: scripts/security-audit.js (Line: 159)
+- Recommended fix: Remove from code, rotate credentials, and load via secure server-side environment management.
+## High
+
+#### Unsafe HTML rendering pattern
+- Severity: high
+- Confidence: medium
+- Status: probable
+- Evidence: dangerouslySetInnerHTML found
+- Impact: Potential XSS if fed untrusted content
+- Files: scripts/security-audit.js (Line: 126)
+- Recommended fix: Sanitize content and restrict sources before rendering HTML.
+#### Server boundary without obvious input validation
+- Severity: high
+- Confidence: low
+- Status: manual-review
+- Evidence: Request input usage
+- Impact: Malformed or hostile input may reach sensitive logic
+- Files: scripts/security-audit.js (Line: 144)
+- Recommended fix: Validate request data at the server boundary with a schema.
+#### LLM endpoint with prompt injection exposure
+- Severity: high
+- Confidence: medium
+- Status: probable
+- Evidence: LLM SDK call
+- Impact: Model behavior may be steered to reveal or misuse privileged instructions and tools
+- Files: scripts/security-audit.js (Line: 149)
+- Recommended fix: Separate system instructions, validate tool args, add allowlists, and sanitize user content.
+#### Potential SSRF pattern
+- Severity: high
+- Confidence: low
+- Status: manual-review
+- Evidence: Server-side outbound request
+- Impact: Attacker may force calls to internal or unintended hosts
+- Files: scripts/security-audit.js (Line: 154)
+- Recommended fix: Enforce host allowlists and validate URLs before outbound requests.
+## Medium
+
+_No findings._
+
+## Low
+
+_No findings._
+
+## Manual review checklist
+
+- Verify authz on admin and privileged routes.
+- Verify whether raw queries are parameterized upstream.
+- Verify whether shared sanitization utilities are applied before dangerous HTML or LLM calls.
+- Verify token handling and public env variables in client bundles.
+