Shared product memory for Codex work.

Code context is necessary, but not enough. The agent needs to know the product promise, the current feature goal, what is out of scope, what the founder has already decided, and what would make the change acceptable.

RAAV packages that context before Codex starts. The result is less drift between a technically valid patch and the product the founder meant to build.

Codex often works in a shared repository where another agent or human may also be editing. Without a coordination layer, overlapping edits become a merge problem and a product management problem at the same time.

RAAV lets Codex claim a task, files, branch, and worktree lane before editing. That creates a visible record of who is working where and gives other agents a chance to avoid conflicts.

A good code change still needs a durable explanation. What changed? Which files matter? What verification passed? What risk remains? What should the founder review before the next agent continues?

RAAV makes Codex submit those details into the ledger. The next Codex session can read the result instead of reverse-engineering intent from commits and terminal logs.