frozenpepper/deepseek-and-destroy ? reverse-engineered prompt

I want you to build a Python based coding workflow helper called DeepSeek and Destroy that helps a premium coding agent stay in control while cheaper worker agents do the heavy lifting.

It should support launching bounded tasks, tracking each attempt in its own folder, collecting worker reports and logs, and checking objective evidence like file changes, hashes, lifecycle state, and write rules before anything is accepted. The main idea is that the parent agent makes the decisions, while worker agents can implement, fix, review, verify, recover, and audit work in a controlled way.

Please make it work as a reusable skill or harness that can be copied into a coding agent setup, with clear prompts, role definitions, task handoff files, and recovery for interrupted or stale attempts. Keep the workflow read only where it should be, and make sure fresh review and verification happen after changes. If you need to, look up the current docs for the agent platforms it mentions.