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

Report privately via one of:

• GitHub Security Advisory (preferred): Open a draft advisory
• Email: security@dxfrontier.com

Include:

• A description of the issue and its impact
• Steps to reproduce
• Affected version(s) (pair-claude --version once available, or commit SHA)
• Your environment (OS, Go version, Tailscale yes/no)

We'll acknowledge within 5 business days and aim to release a fix within 30 days for confirmed issues.

Pair-Claude is designed for trusted networks — typically a Tailscale tailnet shared between teammates. The trust boundary is your network: anyone who can reach the listening port is treated as authorized to attempt to use a token.

In scope (please report):

• Token disclosure via timing attacks, log files, or response leaks
• Authentication bypass on /api/* endpoints
• Cross-site scripting in the embedded viewer
• Command injection through input parsing or claude args
• Memory unsafety in the PTY layer
• Path traversal in the embedded static-file server
• Denial of service that affects the host beyond the wrapper itself

Out of scope (by design — see README §Security Model):

• No TLS. Tailscale provides end-to-end encryption. Running Pair-Claude on a public network without a TLS terminator in front is not a supported configuration.
• Read-write by default. \share produces a token that allows typing into the live session. This is intentional for pair programming. Use \share --readonly for observers.
• Shared token, not per-user. All viewers with the URL share the same token. Pair-Claude does not authenticate individual users.
• 8-hour token lifetime. Tokens persist until expiry, \revoke, or process exit.
• Trusting the network. If an attacker is already on your tailnet, they are already inside the trust boundary.

We currently maintain only the latest minor release. Once 1.0 ships, this table will be updated.