This policy covers the higson Claude Code plugin published in this repository (the skill and the MCP server configuration). It does not cover the Higson product/server itself — report those through your usual Higson support channel.

Please report vulnerabilities privately — do not open a public issue.

Use GitHub's private reporting: go to the Security tab of this repository → Report a vulnerability (GitHub Security Advisories). We aim to acknowledge a report within a few business days and will keep you updated on the fix.

Please include: affected version, a description, and steps to reproduce or a proof of concept.

Things to know when using this plugin:

• Your integration token is marked sensitive and stored in your OS keychain — or, where no keychain is available, in Claude Code's local secret store (~/.claude/.credentials.json, file mode 0600). It is not written to settings.json. The plugin's .mcp.json contains only a ${user_config.token} placeholder — no secret is committed to this repo.
• Use a least-privilege token, not an admin token. What the plugin can do over MCP equals what the token's user can do. An admin token can, among other things, reject or publish other users' work sessions.
• The plugin talks to the Studio instance you configure (studio_url). Your data is sent to that instance, not to the plugin authors.
• Nothing reaches the runtime until you publish. The bundled skill reads before it writes, asks for consent before writing, and never publishes or rejects sessions without explicit confirmation.