Teach your AI reviewer a rule once — it remembers it everywhere
Save a one-off review instruction as durable, org-wide guidance and the AI reviewer applies it to every future review, across every workspace. InfraDots distills your note into a clean rule, you confirm it, and you can remove any guidance from settings.
Anyone who has reviewed infrastructure changes for a team has had the same
conversation more than once. "We always tag production resources with a
cost-center." "A public S3 bucket is never a finding worth blocking — that one
is intentional." "Use for_each, not count." You explain it on one pull
request, and a month later you're explaining it again on another.
Your AI reviewer had the same blind spot: it reviewed every change with fresh eyes and no memory of what your team had already decided. Now it remembers.
What's new
When the agent reviews a change and there's something to implement, the Implement changes dialog now has two separate fields:
- Instructions for this implementation — one-shot details that steer the current change only, and are never saved.
- Remember as guidance for future reviews — a durable rule you want the reviewer to keep.
Whatever you write in that second field becomes a piece of durable reviewer guidance for your whole organization. From that point on, every future review — in any workspace — gets it in context before it runs. It even takes effect on the change you're implementing right now: InfraDots saves the rule just before the agent runs, so this run already knows it. Your one-shot instructions stay one-shot; only the guidance is saved.
You stay in control of what gets saved:
- InfraDots distills it. Your in-the-moment note ("also tag this prod bucket
with cost-center=payments") is generalized into one clean, reusable rule
("Tag production resources with a
cost-centertag.") and sorted into a category — style, security, policy, naming, false-positive, or context. - You confirm before it sticks. The draft opens in an editable dialog. Tweak the wording, change the category, or cancel. Nothing is saved until you say so.
- It's org-wide on purpose. Guidance applies across every workspace and team, so a rule one engineer teaches doesn't have to be re-taught by the next.
Why it matters
- Stop repeating yourself. The reviewer accumulates your team's actual standards instead of rediscovering them one PR at a time.
- Kill the recurring false positive. If the agent keeps flagging something your team has deliberately accepted, teach it once and it stops.
- Consistency across the org. New workspaces inherit the same hard-won conventions from day one — no per-repo onboarding for the reviewer.
This is the concrete version of the "continuous learning" IDP Agents were always meant to do: it learns from the feedback you're already giving.
Guidance never overrides your guardrails
Reviewer guidance is advisory. It shapes how the agent reviews — what it flags, what it lets pass, the conventions it expects — but it can never override InfraDots' destructive-change or security checks. You can't accidentally teach the agent to wave through a resource deletion.
Manage it in one place
Every saved rule lives under Organization → Agents → Reviewer Guidance — a simple list of what your reviewer currently knows, grouped by category. When a rule is out of date, remove it and the agent stops applying it on the next run.
Turning it on
Nothing to flip. Wherever you already use the agent to implement a reviewed change, the Remember this option is right there in the dialog. See the Reviewer Guidance docs for the full walkthrough.