SathiaAI/graphsmith ? reverse-engineered prompt

Reverse engineered prompt

Build me a small JavaScript project called GraphSmith that helps an AI agent run multi step workflows without losing progress, repeating work, or making up details about the codebase.

I want the experience to be simple, first ask for the job in plain English, then show a clear plan before anything runs, then generate a workflow that can be killed and resumed from the last checkpoint. Each step should save progress, have capped retries, and write a clear log so I can see what happened. If a step would cause a duplicate side effect, stop it instead of guessing. If the agent is talking about my own repo, it should cite the real file it got that from, or clearly say it is guessing.

Please include a way to test failure recovery by intentionally stopping a run and proving it resumes correctly. Make it work as a skill that can be used in common coding agents like Cursor, Claude Code, Codex, Gemini CLI, and Copilot. If you need to look up current docs online, go ahead.

Are you gonna build this?

make sure you review the code using coderabbit

Try freeSponsored — opens CodeRabbit in a new tab