LearnHarness stores API keys, sign-in tokens, chats, and learning records on your machine, and its agent reads files and fetches web content on your behalf. If you find a way to abuse any of that, please tell us privately so a fix can ship before the details are public.

Do not open a public issue, pull request, or Discord post for a security problem.

Report it through GitHub's private vulnerability reporting: Security → Report a vulnerability.

If that page is unavailable, message a maintainer on Discord and ask for a private channel. Do not include any details in that first message.

A useful report includes:

• the LearnHarness version and your OS;
• what an attacker can do and what they need first (a web page the agent reads, a file the learner opens, network access, and so on);
• steps to reproduce, or a proof of concept;
• any fix or mitigation you have in mind.

These are targets, not guarantees:

The fix is developed privately in a GitHub security advisory. We will keep you updated, agree on a disclosure date with you, and credit you in the published advisory unless you prefer to stay anonymous. A CVE can be requested through GitHub when one is warranted.

Only the latest release receives security fixes. The desktop app updates itself, so fixes reach users through a new release. If you run from source, update to the latest main or the latest release tag.

In scope:

• the desktop app (Electron shell, window, and IPC);
• the local API server, whether started by the desktop app or run from source;
• the agent and its tools, including prompt injection from web pages, PDFs, or other content that leads to reading, changing, or sending your data;
• rendered content such as demos, figures, and markdown;
• the update and release pipeline.

Out of scope:

• attacks that already require running code as your OS user;
• bugs in third-party model or service providers themselves (report those to the provider);
• automated scanner output without a demonstrated impact on LearnHarness.