Only the latest release receives security fixes. Please update before reporting.

Please do not open a public issue. Report it privately through GitHub Security Advisories.

Include what you found, how to reproduce it and the impact you expect. You will get an answer within a few days; once a fix is released you will be credited in the advisory unless you prefer otherwise.

AgentBoard runs CLI agents in real terminals on your computer, so these areas matter most:

• The agent bridge (src/main/bridge/, scripts/agentboard-mcp.js): a local socket protected by a random token. Anything that lets another local user or a web page talk to it, bypass the token, or write to a plain shell card is in scope.
• Renderer isolation (src/preload/, sandboxed window): any way for terminal output or board content to run code with Node privileges.
• The phone remote (src/main/remote/): bypassing the token, the rate limit or the rule that dangerous requests are approved only at the computer.
• Secrets (src/main/secrets/): reading a secret from the window or from an agent, or a secret appearing in output or logs.
• Approval rules (src/renderer/watchers/ApprovalPolicy.js): a dangerous request that is approved in bulk, from the phone or by an automatic rule.
• Saved data (board.json, secrets.json, audit.log): they should only be readable by your user.
• Worktrees and git commands run on your behalf.

Problems in the agent CLIs themselves (Claude Code, Codex, Gemini CLI…) should be reported to their vendors.