The project is currently in Public Preview. Security fixes are applied to the latest commit on main; older commits and unreleased snapshots are not maintained as separate support lines.

Use the repository's Security tab to submit a private GitHub Security Advisory. Do not open a public issue for a suspected vulnerability.

Include a minimal reproduction, affected component, impact, and suggested mitigation if known. Remove all real Bot credentials, model credentials, employee or conversation identifiers, message contents, media URLs, and local filesystem paths.

Maintainers will acknowledge a valid report as soon as practical, coordinate a fix and disclosure timeline, and credit reporters who wish to be credited.

Operators are responsible for:

• keeping Bot, Kernel, and provider credentials outside the repository;
• using scoped direct and group allowlists;
• restricting .env, SQLite, media temp, and spool permissions;
• restricting WECOM_MEDIA_OUTPUT_ROOTS to dedicated output directories;
• leaving raw SDK message and Adapter stderr logging disabled except during a short, controlled diagnostic window;
• granting Adapter child processes only the minimum required environment variables through their explicit allowlists;
• backing up the private data directory before an upgrade and respecting SQLite schema-version failures instead of bypassing them;
• running exactly one Gateway Kernel for a Bot identity;
• reviewing write-tool approvals and rotating credentials after suspected exposure.

The project does not provide a security warranty. See LICENSE.