How the Luna Access Review Agent works
The Luna Access Review Agent chases reviewers, proposes submissions and reassignments, and reports progress for access review cycles whose owners delegated chasing to Luna. It is one of the agents we build and ship with Ploy, alongside the Flow Error Analyzer. For how these pre-built agents differ from the playbooks you run yourself, see What are built-in agents?
When it acts
The agent is disabled by default and shows the tagline "Helps get access reviews completed on time." It is created when a campaign owner first delegates chasing to Luna, and until then its status shows "Enables when needed".
Once it exists, it acts on a cycle only when all three of these hold:
The cycle status is in progress.
The cycle's start date is today or earlier (UTC).
The cycle has delegated chasing to Luna (Luna assist is enabled).
It does not act on drafts, cycles that start in the future, or cycles where delegation is disabled. If it is pointed at a cycle that has not delegated, it refuses the request with the message "this campaign has not delegated chasing to Luna" and drops it.
What it does
The agent's instructions give it four duties: chase, decide, report, and record.
Chase: it works out what remains in each review and nudges reviewers to keep going, following the campaign's cadence.
Decisions: it proposes submitting completed reviews to the reviewer, and, when permitted, proposes reassigning reviewers who have gone silent.
Report: it gives progress updates to the campaign owner or a configured report channel, daily at 08:00 UTC.
Record: it annotates every nudge, question, action, and report it makes, so the trail is auditable.
Chase cadence and quiet rules
By default the agent nudges reviewers every 3 days. Once 70 percent of the review window has elapsed, it nudges daily, unless the campaign's process instructions say otherwise.
Several quiet rules keep the nudging from becoming noise:
A reviewer who has not responded has a silence threshold of 7 days before the agent treats them as silent.
The agent sends at most one proactive outbound message per person per day, across all cycles.
If a person has several review sets, the agent combines them into one nudge.
It does not message a reviewer on a day Ploy has already contacted them, or when the next Ploy reminder is due today or tomorrow.
How the agent works alongside your notifications
Delegating chasing to Luna does not turn anything off. Your reminders and escalation tiers run as set, and Luna chases on her own cadence alongside them. Ploy does not suppress its own notifications to make room for her, and she does not replace any of them.
Escalation tiers keep firing on delegated cycles. Where a reassign tier exists, Luna leaves moving reviewers to the tier.
Scheduled alerts, such as the overdue and incomplete-review notifications, still go to the channel or member you configured. They are separate from Luna's reviewer-facing nudges.
Manual reminders still work: you can use Send Reminders on a running cycle whenever you want, and a reminder Ploy sends counts as a chase for Luna, so she nudges that reviewer next only once her cadence has passed since it.
If you have turned a notification type off in Settings, Ploy does not send it, and Luna does not count it as a contact when timing her own messages.
Our recommendation when the agent is live: let Luna manage the chasing. Your existing reminder and escalation settings can stay as they are. If you would rather reviewers hear only from your escalation tiers, say so in the campaign's process instructions and Luna leaves the chasing to them. None of this holds back decisions and replies: she always answers a reviewer's question and proposes a submission without delay.
For what each setting controls, see Access review notifications and reminders.
What stays with humans
Submission decisions stay with the reviewer. The agent can propose submitting a completed review, but the reviewer must confirm it themselves using the decision controls: typing yes or no in chat does not confirm it, and an administrator cannot confirm on their behalf. Only one submission proposal is open per reviewer at a time in a cycle.
Reassignment proposals go to the campaign owner or someone authorized to reassign reviewers, and the agent proposes them only when the owner has authorized it or the campaign's process instructions permit it. If the owner rejects a reassignment candidate, the agent does not reuse that candidate for the same set.
What you see
The agent's work shows up in the places you already use for reviews: nudges to reviewers, submission and reassignment proposals, daily progress reports, replies to reviewer questions in the review conversation, and annotations on campaign cycles, reviews, and account sets. These records are not Luna artifacts.
This agent is not the same as Luna's account-level recommendations in access reviews, which suggest an outcome (auto-approve, auto-reject, or needs review) for each account a reviewer is checking. That is a separate feature; see Luna recommendations in access reviews.