Answers
Should team knowledge for AI live in versioned files, or in curated memory?
Files in the repository — a CLAUDE.md, a folder of skills, notes committed next to the code — are where most teams put what their AI should know, and that is where the trouble starts: whatever lands in them governs every later session, and nobody sees it arrive or approves it — an assistant's guess, a status that went stale, a decision the team never knew was made. A file also cannot say which rule is in force today, who decided it, or whether the model actually received it. Arroway holds what should hold: written by people and their assistants, approved by a person, ordered by the task, and delivered with a count of what fit. The file can stay empty.
Last updated August 25, 2026
A file is corrected by hand, and only after someone notices
A versioned file is as current as the last person who remembered to edit it. Nothing in it tells you a line has been overtaken: a superseded rule reads exactly like a live one, and it keeps being followed until a human writes the correction. In Arroway each memory carries the condition that ends it, a replaced memory points at what replaced it, and a guess waits for a person instead of governing quietly.
A repository is a boundary, and half the team is on the other side of it
Knowledge in a repository reaches whoever works in that repository. The pricing decision, the rule about what may be sent to a client, the reason a supplier was dropped — those belong to the same team, have no repository to live in, and are often made by people who never open one. Arroway is organised by project, not by codebase, so a decision written by someone who never touches git reaches the assistant of someone who does.
Loading a file is not the same as knowing it arrived
A file is text placed in front of the model, and the file cannot say how much of it fit. Arroway's read is ordered by the task the assistant declares and comes back saying what it delivered and what did not fit — so “the AI has the rule” stops being an assumption and becomes something you can look at. That is the difference between context that was sent and context that landed.
What it looks like in practice
A team writes its release rule into the repository's context file. Two months later the rule changes in a meeting, and nobody touches the file — why would they; the meeting was not about files. Every assistant keeps quoting the old rule, correctly, from a file nobody thought was stale. Nobody forgot anything. A file simply has no way to say it was overtaken.
Questions people ask about this
- Should we delete our context files, then?
- Empty them. Arroway's recommendation is that nothing stays in an instruction file: whatever is written there governs every session with nobody seeing or approving it, which is how junk, stale state and decisions nobody knew about pile up. What still holds moves to Arroway, where a person sees it and approves it before it governs anything. What is noise or out of date is simply deleted.
- Why not just commit the decisions too, with dates?
- Teams do, and a date does not fix it. What a committed file still cannot do is tell an assistant's guess from a decision a person approved, say which of two contradicting lines is in force today, or report how much of the file the model actually received. Those are the three failures this page is about, and none of them is solved by writing more carefully.
- What does a person actually have to do to keep this current?
- Review, not writing. Assistants record what they learned as proposals; a person approves, edits or discards them on a short list, and only what a person approved governs the next read. The work is judging that list once in a while — minutes, not a maintenance job — and skipping a week breaks nothing.
Where this is verifiable
Product documentation on this site (How it works, Install) and the sanctioned product spec in the repository. The description of files-in-a-repository is about the practice, not about any vendor's product, and every property named here is one a plain versioned file has: no expiry, no sanction step, no delivery receipt. This page makes no claim about the built-in memory of any specific AI tool.
https://www.arroway.app/en/answers/context-files-vs-curated-memory