A PR review bot that knows your decisions

CI and automationTeam with an agent reviewing every pull request

The review bot flagged 'prices as integers' and 'no Redux' on every pull request. People started ignoring it.

How do I stop our AI pull request reviewer flagging choices the team already settled?

Run the review job's agent with ArcRoom so team decisions come in through recall. Settled choices, like cents pricing or no Redux, are no longer flagged, and comments focus on real problems, so people read them again. Only decisions that were recorded are respected.

Also asked as: People ignore our review bot because it repeats the same comments. How do we fix it? · How do I give an AI code reviewer our team's past decisions? · Can an AI PR review bot skip things we already agreed on?

The work

The review job runs the agent with ArcRoom.

ArcRoom keeps
  • Decision
  • Recall
The next agent gets

[review bot] 2 comments. Skipped 3 that match team decisions (cents pricing, no Redux, tRPC only).

Step by step

  1. The review job runs the agent with ArcRoom.
  2. Team decisions come in through recall, so settled choices aren't flagged.
  3. Comments focus on real problems, and people read them again.

What the agent sees

[review bot] 2 comments. Skipped 3 that match team decisions (cents pricing, no Redux, tRPC only).

Compared with today's tools

Decisions in the review prompt or AGENTS.md / CLAUDE.md

The team lists settled decisions in CLAUDE.md or the review workflow's prompt, and the bot reads them on every run.

Setup
30 min
Ongoing
Add each new decision by hand, ~5 min

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 (npx arcroom init)
Ongoing
None

The file is loaded whole on every run and goes stale quietly. ArcRoom recalls only the relevant decisions and replaces overturned ones. For a short list, the file works fine.

Where it stops

Only recorded decisions are respected.

Try it in your project

Works with the agents your team already uses, and any agent over MCP or the HTTP API.

More in ci and automation