AdityaGaur77/Agent-harness-26 ? reverse-engineered prompt

Build me a privacy erasure agent called Blast Radius.

I want it to help delete everything tied to a customer, but in a careful way. It should first inspect the database, find the rows connected through foreign keys, figure out what would be deleted, and then rehearse the plan on a shadow copy so it can see the real impact before doing anything irreversible. If a naive delete would remove data that has to be kept for legal reasons, it should revise the plan so it hard deletes personal data and anonymises anything that must remain. After that, it should rehearse again and only then stop for human approval before running the real delete.

Please make it work with the existing Postgres and Redis setup, expose the MCP server with auth, and include the agent provisioning flow. If there’s any missing piece, look up the current docs online and wire it up cleanly.