Only the latest minor release receives security fixes.

Please do not file a public GitHub issue for security problems.

Open a private security advisory with:

1. A description of the vulnerability and its impact.
2. Steps to reproduce, or a proof of concept.
3. The aidit version (aidit --version), Node.js version, and OS.

You should get an initial response within 72 hours. If the issue is confirmed, we will work with you on a fix and a coordinated disclosure timeline, and credit you in the advisory unless you prefer otherwise.

aidit edits files that tell AI agents which programs to execute (MCP server command entries). Anyone who can write to those files can run code as you, so the local API is treated as security-sensitive.

What aidit does to protect it:

• Localhost only. The server binds to 127.0.0.1, never 0.0.0.0, so it is not reachable from your network.
• Host and Origin checks. Requests whose Host header is not a loopback address are rejected (DNS-rebinding protection), and state-changing requests and WebSocket connections from foreign origins are refused. No permissive CORS headers are sent, so a web page you visit cannot drive the API from your browser.
• Optional token auth. aidit start --token requires a bearer token on every API request.
• Safe writes. Every write is validated, the previous file is backed up under ~/.config/aidit/backups/, and the new content is written atomically (temp file + rename).
• Your permissions only. aidit never asks for sudo and only touches files your user can already read and write.
• No telemetry. aidit makes no outbound network requests.

In scope: anything that lets a remote party (a website, another machine on the network) read or modify config files through aidit, path traversal outside registered paths, command injection, and bypasses of the backup/validation pipeline.

Out of scope: attacks that require an attacker who can already run code as your user, and vulnerabilities in the agents whose configs aidit edits.