If you discover a security vulnerability in Human Markdown, please report it privately via email:

security@purecontext.dev

Do not open a public GitHub issue for security vulnerabilities.

You should receive an acknowledgment within 48 hours. We aim to release a fix within 7 days for critical issues.

Human Markdown is a local-only VSCode extension. It processes untrusted input (user-authored markdown files) and renders it in a VSCode webview. There is no network communication, no telemetry, and no server-side component.

The VSCode webview is the primary security boundary. It runs in a sandboxed iframe with:

• No filesystem access
• No Node.js runtime
• No network access
• No access to other extensions

The webview communicates with the extension host exclusively through postMessage. The message protocol accepts only four message types: edit, open-link, save-state, and ready.

Every webview sets a strict CSP:

• default-src 'none'
• script-src uses nonces — no inline scripts
• style-src allows inline styles for theme injection
• img-src allows HTTPS and data URIs
• No connect-src — no outbound network from the webview

• Mermaid runs with securityLevel: 'strict'
• KaTeX runs with strict: true
• Shiki produces sanitized HTML from syntax highlighting
• All three are loaded as separate IIFE bundles with CSP nonces

• GitHub Dependabot monitors for known vulnerabilities
• pnpm audit runs in CI on every push
• Lock file (pnpm-lock.yaml) is committed for reproducible builds

The following are in scope for security reports:

• XSS or code execution via crafted markdown files
• CSP bypasses in the webview
• Vulnerabilities in direct dependencies
• Path traversal via image or link handling
• Extension host compromise via message passing

The following are out of scope:

• Vulnerabilities in VSCode itself
• Issues requiring physical access to the machine
• Social engineering attacks