Keep a changelog from what shipped

Every time the person ships, put one row for that release into the list they keep their work in: read their newest release (from the release program or git ref they name) with the subjects of the commits since the release before it, make the row with their own mapping, and write it only when the list lacks it, so a row someone already edited stays as they left it. The AI writes no words people read and never marks a row public by itself.

What it asks for

  • work (Needed, words): A folder for the sender's plan, result and status files
  • roots (Optional, words): The --root NAME=DIR arguments the mapping needs, if any
  • staged (Needed, words): The staged file of releases the reader keeps
  • mapping (Needed, words): The person's build-records-to-write mapping that makes one row of their list from each staged release
  • payload (Needed, words): Where the built payload is written
  • settings (Needed, words): The read-what-shipped settings file: where the newest release is read and which keys hold its commit, name and time
  • repository (Needed, words): The person's repository folder
  • door_config (Needed, words): The sender config for the person's door, with keep_held true

The steps your AI walks

  1. Note the newest release, if it is new
  2. Make the rows from the staged releases
  3. Write only the rows the list lacks

Use it from your AI

Connect the AI you already use to Trillion once. Then ask in your own words; your AI finds this workflow by what it does and runs it.

Say to your AI: “Use Trillion to every time the person ships, put one row for that release into the list they keep their work in: read their newest release (from the release program or git ref they name) with the subjects of the commits since the release before it, make the row with their own mapping, and write it only when the list lacks it, so a row someone already edited stays as they left it. The AI writes no words people read and never marks a row public by itself”