First, determine whether the user needs to deploy a new instance or connect to an existing one. The Zhixu open-source repository is at https://github.com/dmety/Personal-knowledge-base. This skill package is guidance and workflow only; it does not contain callable shared MCPs or provide the author's site resources.
For deployment, follow the Sites deployment steps. If available Sites plugins exist, use their skills and native workflows; otherwise, provide steps to install the Sites plugin and open the repository, but do not claim resources have been created. Operate only within the user's own cloned directory and account. Preserve existing instance IDs; if a creation times out, first query to confirm, and do not create duplicates. Do not copy another user's project ID, connection address, identity header, or credentials.
For connection, have Sites query the target instance via get_site(include_mcp_connection=true) and use the returned personal plugin installation entry point. Alternatively, instruct the user to navigate to Plugins > Personal > Created by you, select their instance, and click Install / Connect. Do not change a platform-hosted connection to a hand-written local MCP or hardcoded remote address.
After confirming that the data tools originate from the target instance, use the read-only search_knowledge to validate. An empty result is a valid response; if no actual successful call has been made, explicitly state that the connection is still pending verification. When multiple connections are available, first confirm which one is the target for this session. Do not create drafts to test connections, and do not read unrelated personal data.
Once complete, use $knowledge-workflow for collection and organization. AI should only propose content for review and organizational suggestions; final review is completed on the user's site.