State in the repository (.agents/)

The idea

A repository's .agents/ folder holds rules, memory notes, task checks and summaries. It is shared through git and read back in.

The problem
Agent knowledge stayed inside one tool, and the team lost it.
Why it matters
The team reviews and shares agent state like code.
Where it stands
In production as of

A step-by-step replay of this idea is not published yet. The library adds one when a recorded run shows it.

  • Rules as checks

    A team rule that can be checked gets a checker for a pattern, a name, a structure, a command or a benchmark. The checker runs on the change before a pull request, on each push, before the worker finishes the task, and during verification.

  • Project memory scope

    A memory can belong to one repository, and only work that names that repository recalls it. The visibility rules do not change.

  • Timeline: automatic cross-system activity feed

    An automatic feed of what one identity or Everyone did across GitHub, Jira, Slack and other connected systems; an identity can be a person, an outside agent, an AI worker or an automation.

  • Rulebook mined from history

    Commits, PRs, reviews, tickets, tests, CI and code are mined into rule candidates with evidence and scope. Accepted rules become checks, tested first against past changes. A rule that catches nothing in 90 days goes to its author to retire or keep.

  • Context contract per role

    Each model call carries only its role's layers, in a fixed order. People's messages enter as task state with their ids. Budgets are kept by folding and ranking, never by cutting the prefix.

Turn the numbers into shipped work.

Crewshal runs these choices for you: a persistent AI worker with memory and rules, on your Claude and Codex subscriptions.