One prompt, three jobs. We start from an empty folder and describe how a sales pipeline can live in plain folders: one folder per stage, numbered so they sit in order, one folder per prospect inside its stage, and moving a deal means moving the folder. Each prospect gets an activity file with an overview and an append-only log, plus messages and documents folders that appear only when needed. Around it: an inbox for raw inbound, a processed folder for filed originals, and a reroute folder for anything that isn't a sales item. Company docs, a prospect seed, and past follow-up emails go in reference. Then, in the same prompt, three asks in order: set up the stage folders, populate them from the seed (a longhand pipeline becomes eight prospect folders spread across six stages, each with an activity file seeded from its history), and write the AGENTS.md that makes it all self-describing. Ninety seconds in, the pipeline lives on disk instead of in your inbox and your head. And you've seen that an agent can hold several jobs in one well-ordered prompt; you don't have to feed it one step at a time.
What's covered:
- The file-based CRM: stage folders, prospect folders, and the folder is the stage
- The activity file: header plus append-only log
- inbox, processed, and reroute, and what each is for
- Reference material for the pipeline, including your own sales voice
- Writing the AGENTS.md for a repo with rules
- Building the prospect folders from a seed doc
- Packing several jobs into one prompt: set it up, fill it, document it
- No app, no database, no server, no login