Use GitHub's "Report a vulnerability" link on the Security tab of this repository. Please do not file a public issue for security reports.

We aim to acknowledge reports within 5 working days.

In scope:

• Code in this repository (src/, dist/).
• The local HTTP servers (home-server, plan-server) — both bind to 127.0.0.1. Reports of cross-origin escalation, path traversal, or command injection through their endpoints are appreciated.
• Subprocess spawning (runProc) — argument-injection or unintended path traversal via user input.

Out of scope (not bugs in this project):

• Bugs or vulnerabilities in upstream dependencies (ffmpeg, whisper.cpp, @modelcontextprotocol/sdk). Please report those upstream.
• Misuse of the tool against content the user does not own. See NOTICE.md for the project's stance on responsible use.
• Issues caused by running the tool on untrusted videos with non-default configurations (e.g. a manually edited plan.json containing malicious ffmpeg filter expressions). The project assumes plans are produced by the user or their AI host, not by attackers.

aicw-video is a single-user, local tool. It does not host content, does not accept incoming network connections from anywhere except 127.0.0.1, and does not send telemetry. The threat model assumes:

1. The user trusts the videos they import.
2. The user trusts the AI host they connect (Claude, OpenClaw, etc.).
3. The user is the only person with shell access to their machine.

If your environment violates any of those assumptions, the tool may not fit your security needs.