MangoStudio
PTEN
StarGet started
Operations / Security Policy
Source on GitHub

Security Policy

Reporting a Vulnerability

If you discover a security vulnerability in MangoStudio, please report it privately rather than opening a public issue.

Email: Create a private security advisory through GitHub’s Security tab, or email the maintainers directly.

Response SLA: We aim to acknowledge reports within 48 hours and provide a fix timeline within 5 business days.

Supported Versions

Only the latest commit on the main branch receives security fixes. We do not maintain backport branches for older versions.

Security Model

Authentication

  • Better Auth with email/password strategy.
  • Session-based authentication via HTTP-only cookies.
  • Sessions are validated on every protected API request via requireAuth middleware.
  • User passwords are hashed using bcrypt (configured via Better Auth).

API Key Storage

API keys (Gemini, OpenAI, Anthropic, DeepSeek) support three persistence methods, in priority order:

  1. Environment variables (~/.mango/.env) — Highest priority, never committed.
  2. config.toml (~/.mango/config.toml) — Flat file, never committed.
  3. OS Secret Store (Bun.secrets) — Native OS-level secure storage. Recommended for maximum security.

Keys are never exposed to the frontend. The API resolves secrets server-side and never includes them in responses.

Rate Limiting

In-memory rate limiter with configurable window and max requests per window. Protects against brute-force attacks and API abuse. Can be configured to trust proxy headers when deployed behind a reverse proxy.

Input Validation

  • All API request bodies are validated via TypeBox schemas before processing.
  • File uploads are validated for MIME type, magic bytes, size, and UTF-8 correctness.
  • Path traversal protection on all file operations (upload storage, generated image storage, attachment resolution).
  • URL validation on provider base URLs (rejects private/loopback addresses).

Data Protection

  • The database file (database.sqlite) contains user data. Restrict filesystem access to the process user.
  • No PII is logged. Passwords, tokens, and raw authentication payloads are excluded from all log output.
  • Error messages returned to clients do not contain sensitive data, stack traces, or internal paths.

Build Security

  • bun build --compile produces standalone binaries without exposing source code.
  • Frontend assets are minified and tree-shaken.
  • Environment variables are not bundled into the binary.

Automation Security Gates

Pull requests are checked by these GitHub security workflows before merge:

  • CodeQL (.github/workflows/codeql.yml) runs on pull requests to main, pushes to main, a weekly schedule, and manual dispatch. It uses CodeQL advanced setup for JavaScript/TypeScript and Rust (on Linux and on Windows, so cfg(windows) code is scanned too) with explicit no-build extraction and the security-extended query suite. The workflow uploads SARIF with a stable per-language category so alerts are visible in the Security tab and in the PR Code scanning results / CodeQL check.
  • Dependency Review (.github/workflows/dependency-review.yml) runs on pull requests to main and blocks newly introduced moderate-or-worse vulnerable dependencies from manifest or lockfile changes. License checks are intentionally disabled.
  • Dependency Audit (.github/workflows/dependency-audit.yml) audits every package in bun.lock for moderate-or-worse advisories on a daily schedule and on pull requests that change a manifest or the lockfile. Unlike Dependency Review, it also flags advisories published after a version was locked.

When GitHub posts an Advanced Security setup comment on a PR, use it as a pointer to the Security tab, not as the review process itself. Maintainers should review the CodeQL analysis job, the Code scanning results / CodeQL check, and the Dependency Review check before merging security-sensitive changes.

What NOT to Report

  • Missing GPG signatures on commits (this is a contribution guideline, not a security boundary).
  • Unpopulated .env files in the repository (these are .gitignored by default).
  • Rate limiting configuration suggestions (these are configuration concerns, not vulnerabilities).
  • Theoretical attacks that require filesystem access to the server (if an attacker has filesystem access, the security model is already compromised).
← PreviousDeployment
esc
↵ openesc closeMangoStudio