create-system-tool

Create a TypeScript file in plugin/environment/src/system_tools, decorate the class with @glyvio_core.SystemTool, and implement CoreSystemTool with an async handle method returning a string. Then add a tool permission to manifest.json, run run_helper.sh, and import the file in the environment entrypoint.

Use a SystemTool when the behavior must be callable by the AI Agent during natural-language chats, such as reading local status or registering records. Use an @Action for public RPC methods or heavy local processing not exposed to the AI.

The compiler exposes request field JSDoc to the LLM, so missing or vague JSDoc comments prevent the agent from building valid requests. Also verify the permission exists in manifest.json, run_helper.sh was executed, and the file is imported so the decorator registers.

Yes, the environment layer can query the customer's local synced PostgreSQL replica using the SyncClient from the sync plugin after declaring the dependency in manifest.json. You must resolve a real environment ID via a query rather than hardcoding one.

The handle method must return a string designed to be readable by an LLM, ending with explicit "Instructions for the LLM" text describing how to present the data. Business failures should be thrown as glyvio_core.GlyvioError rather than caught silently.