Security

Reporting

Please report security issues privately using the GitHub Security Advisories feature:

• Go to our Security Advisories page.
• Explain the vulnerability, the potential impact, and steps to reproduce.
• Include relevant context such as version, OS, shell, runtime, and install method.

By using GitHub's private vulnerability reporting, you help ensure the issue is resolved before it is disclosed publicly. Once a fix is available, we can credit you in the release notes unless you prefer to stay anonymous.

Supported versions

Security fixes go to the latest published release.

What is in scope

• The CLI in packages/cli, and the packages it runs on: ai, agent, tui, durable, protocol, client, server, tools, and session-backend-sqlite. That includes tool execution, credentials, model calls, and terminal rendering.
• Website code that could expose users or misrepresent downloads.
• npm package publishing or release integrity for KnightCode.

What is not

• Issues in upstream dependencies unless KnightCode needs a specific mitigation.
• Provider-side behavior from the model, search, or hosting service you configure.
• Anything that requires an already-compromised local machine.

How KnightCode reduces risk

• Visible tool calls. Reads, edits, searches, and commands show up in the transcript as they run.
• Project trust. A folder's settings, skills, and extensions load only after a trust decision. Trust does not sandbox tool calls.
• Optional classifier gate. When enabled, it can hold a risky shell command or edit for confirmation, or block it in print and JSON modes.
• BYOK model access. KnightCode does not run a hosted model proxy for your code.

What we cannot promise

• KnightCode can read, change, and execute files with your operating-system permissions, and it does not ask before every tool call. Use source control and read the transcript.
• Providers see whatever context you send them. Review their retention and training policies.
• Keep the CLI updated and report suspicious behavior.