Library maintainers and their decisions
Agencies and freelancersCore team of a widely used library
Maintainers kept explaining in reviews: 'we don't add dependencies for this', 'Node 18 support stays until April'. Their own agents forgot too.
How do open-source maintainers get their AI agents to follow project decisions?
Record project decisions in your ArcRoom team, like 'no new runtime dependencies without an RFC', with the reason. Maintainers' agents then follow them when writing and reviewing. Outside contributors don't share the memory unless they're on the team, and public contributors don't get it today.
Also asked as: I keep repeating the same project rules in code reviews. How do I make agents remember them? · How do I give my AI agents an open-source project's policies? · Can AI agents remember why we still support an old Node version?
Maintainers record project decisions in their ArcRoom team.
- Decision
Decision: no new runtime dependencies without an RFC. Why: install size.
Step by step
- Maintainers record project decisions in their ArcRoom team.
- Their agents follow them when writing and reviewing.
- Outside contributors don't share the memory unless they're on the team.
What the agent sees
Decision: no new runtime dependencies without an RFC. Why: install size.Compared with today's tools
CONTRIBUTING.md + AGENTS.md
Maintainers write the policies into repo files, so every contributor and their agent can read them.
- Setup
- 1–2 hours
- Ongoing
- keep the files current as decisions change, ~15 min a month
ArcRoom
Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.
- Setup
- 2 min per maintainer
- Ongoing
- None
Repo files win for outside contributors because everyone can read them; ArcRoom only helps the core team's agents, recalling a decision when it's relevant instead of loading all of them every prompt.
Where it stops
Public contributors don't get the memory today. A public read-only room could come later.
Try it in your project
Works with the agents your team already uses, and any agent over MCP or the HTTP API.