Cards are enriched markdown

A box is your data directory, a card is one file in it, and the agent is the coding agent that reads and writes those files. A card is ordinary markdown, enriched in two specific places. Everything else about it is plain text you can open anywhere.

The header is extended per kind of card. Every card starts with a structured header that the system checks against the card's type. A recipe has a yield and ingredients; a person has aliases and a role; a question has its options and where the answer goes. A box can define new kinds with their own fields, so the header vocabulary grows with what you keep (see dump it in now, shape it later).

The body is markdown with a small baseline of marks. The text below the header is markdown, rendered with a shared vocabulary of marks that any card can use. Three are the baseline because they are cross-cutting: they apply to any kind of card, and the box collects and shows them wherever they appear.

• Quote: exact words, with who said them. This is how your own words survive the agent's rewriting, and how a remark from an email or a chat is kept as it was said.
• Source: where a passage came from, in the box or outside it, and how it was used (verbatim, summarized, inferred). This is how a fact carries its origin.
• Todo: a task noted in place, in whatever card the thought occurred, with a status and optional dates. Every todo view in the box is a live query over these marks, so a task written inside a project note shows up on the plate without being copied anywhere.

Other marks exist for particular kinds of card (a recipe's ingredients and steps, a briefing's purpose and corrections, a capture session's timeline), and a box can add its own. The baseline three are the ones the whole system agrees on.

Why it is built this way: the representation mirrors the idea and durability and provenance. Where the marks matter most: provenance, an inbox for your thoughts, and a project's working memory. The format details are in cards.