Please email security@deconstruct.ai (replace with the actual address before publishing) with a description of the issue and a reproducer. Do not file a public issue for security bugs.

Deconstruct.ai accepts user-uploaded project archives, parses their contents, sends them to the Google Gemini API, and renders the structured response in the browser.

The following controls are in place:

• Strict input validation: file type, MIME, and size caps; zip-bomb guard; path-traversal rejection; symlink-free extraction; per-entry and total byte caps.
• Untrusted-content fences: source code is wrapped in <UNTRUSTED_CODE> blocks and the system prompt explicitly instructs the model to ignore instructions found inside the data.
• Secret redaction: known API-key patterns are removed from the code payload before it leaves the server.
• Prompt-injection regex guard with explicit rejection of suspicious uploads.
• AI output is treated as untrusted: HTML-escaped, length-capped, and validated against an allow-list of file paths and line numbers.
• The Gemini API key is read on the server only via server-only modules; it is never exposed to the client.
• Strict Content-Security-Policy, plus X-Frame-Options, X-Content-Type-Options, Referrer-Policy, Permissions-Policy, COOP/CORP, and HSTS headers via Next.js config and middleware.
• Per-IP rate limiting on /api/analyze.
• Monaco Editor is loaded locally (no remote module fetches) and is configured read-only and DOM-read-only.
• Errors are logged server-side with secret redaction; clients receive only sanitized messages.

• next.config.ts disables x-powered-by and source maps in production.
• productionBrowserSourceMaps: false ensures the original source is not shipped to clients.
• All dependencies are pinned to patched versions. Run npm audit --omit=dev --audit-level=high regularly.

• Vulnerabilities in third-party APIs (Google Gemini).
• Denial of service via network bandwidth, which the hosting provider must mitigate.