Stages that leave a file
Build-along · ~25 min at your own keyboard
Run a long job as stages. Each stage reads one file and writes one file. The agent starts from the file the previous stage left and nothing else. Between stages you read the file and decide whether to continue. This is where you catch mistakes while they are small. If stage two picks the wrong customer, you find out from the stage two file, before stage three builds on it. It also lets you stop for the day and start again tomorrow, because the state lives in files and not in the chat.
Ladder · Level 6The long version: Workflows ↗
Do this
0/7Checkpoints
- ▸You should now see three stages, each with an input file, an output file and a yes or no check.
- ▸You should now see a stage one file marked approved with a date.
- ▸You should now see stage two built from that file alone in a fresh chat.
The three ways this goes wrong
- All the stages ran in one long chat. — Start a fresh chat for each stage and hand over only the file. If it fails, the file is missing something. Add it.
- Your checks say things like looks right. — Rewrite each as a question with a yes or no answer, tied to something you can see in the file.
- You approved every file without reading it. — Read at least the top and the bottom of each one. Mark approved only after you answer the check.
Share what happened
Paste your stage list with input file, output file and check. Then paste the approved stage one file header and what stage two produced from it.
Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].
🔒 Don't paste confidential, personal-health, or payment info.