Cookies on this site

With your OK we use Google Analytics and Google Ads cookies to see which pages and ads bring people here. Nothing loads unless you accept. Privacy policy The LimitRadar app is not affected and still sends no data outside Atlassian.

Privacy policyChange it any time in Cookie settings at the bottom of the page.

Radrly
LimitRadar
Try it free

Documentation

Everything you need to install, configure and use LimitRadar.

Documentation version 1.8, updated 12 November 2026. It describes the LimitRadar code running in production (Forge app version 4.0.0).

On this page

Overview

LimitRadar is an Atlassian Forge app for Jira Cloud. It measures usage of the 10 enforced Jira Cloud configuration limits and the 4 guardrails, forecasts when each space, scheme, workflow or field will reach its limit, lists cleanup candidates with a safety level and the reason, simulates the effect of a cleanup and runs guarded cleanup actions with undo wherever Jira allows it.

New to LimitRadar? Read What is LimitRadar? for a short introduction.

Analysis is read only. Every change to Jira is an explicit command by a Jira administrator (or, on the space settings page, a space administrator), with typed confirmation, a fresh check of the item against Jira and an entry in the actions log.

Requirements

  • Jira Cloud on any plan (Free, Standard, Premium or Enterprise). The limits apply to every plan.
  • Jira administrators (global permission Administer Jira) use the main app under Jira settings → Apps → LimitRadar, the data in the dashboard gadget and the Rovo agent.
  • Space administrators (permission Administer spaces) use the LimitRadar page in their space settings.
  • Rovo is needed only for the Rovo agent. Every other feature works without it.

Install and get started

  1. Install LimitRadar from its Atlassian Marketplace listing, either on the Marketplace website or from Apps → Explore more apps inside Jira. A Jira site administrator approves the requested scopes during installation.
  2. Let the first scan run. Installing the app queues the first scan. It reads your configuration, takes the first snapshot and reads field creation records from the Jira audit log. You can leave the page; progress is kept, and the scan continues in steps until the snapshot is finished.
  3. Open LimitRadar from Jira settings → Apps → LimitRadar. Jira also lists a Get started with LimitRadar page with the same first steps. Until the first snapshot finishes, Overview shows a checklist: start the first snapshot, enter a migration date (or an alert space), and review the cleanup defaults.
  4. Review Overview and Next to block. See which scopes are closest to a limit and check a planned bulk add with What if I add.
  5. Plan your first cleanup. Open Cleanup, choose a candidate type, add candidates, simulate and save a plan. Write mode stays off until you enable it in Settings.

LimitRadar follows Jira's light and dark themes.

Limits and counting rules

LimitRadar measures the 10 enforced limits that Atlassian publishes for Jira Cloud configuration. The catalog ships inside the app as a versioned file with each value, scope, enforcement status, counting rule and source. LimitRadar never downloads limits from the network; a change by Atlassian arrives as a new app version. The guide Jira Cloud limits lists every limit and guardrail with its Atlassian source.

Enforced Jira Cloud configuration limits that LimitRadar measures
LimitValueHow it is counted
L1. Fields per space700Fields in the field schemes (or legacy field configuration schemes) associated with the space. Trashed fields do not count.
L2. Work types per space150Work types in the work type schemes associated with the space.
L3. Options per field20,000Options of a single field, across all its contexts.
L4. Work item security levels per space50Measured per work item security scheme and shown for every space that uses the scheme.
L5. Grants per permission50Grants of one permission in a permission scheme.
L6. Releases per space15,000Archived releases are not intended to count. LimitRadar also shows the count including archived releases.
L7. Workflows per workflow scheme150Distinct workflows in one workflow scheme.
L8. Statuses per workflow200Statuses in one workflow.
L9. Jira components per space10,000Jira components only. Compass components do not count.
L10. Priorities per space100Priorities in the priority scheme assigned to the space.

How enforcement works in Jira

  • Existing configuration above a limit keeps working, and Jira refuses to add more of that entity until usage is reduced. Jira does not delete or change configuration automatically, and people keep creating and updating work items.
  • A bulk action that would exceed a limit is rejected in full. What if I add tells you whether a planned bulk add will pass.
  • Each limit card shows the enforcement status as Atlassian publishes it: Enforced since 2026.03 for fields and work types, and Progressive rollout from 2026.09 for the other eight. Jira does not report whether a progressive limit is already active on your site, so LimitRadar does not claim that it is.

Counting notes

  • Fields per space: LimitRadar detects whether your site uses the new field schemes or the legacy field configuration schemes and counts from the right source. Team-managed spaces count their own fields. Next to the count, LimitRadar shows the count excluding locked fields.
  • Releases: the main number excludes archived releases; the count including archived releases is shown next to it.
  • Security levels and priorities are measured per security scheme and per priority scheme and shown for every space that uses the scheme.
  • If a scan step cannot read the data for a limit, the card says Not available: missing permission or Not counted: the scan step failed. LimitRadar never shows an invented zero.

Guardrails

Guardrails are recommendations published by Atlassian. Jira does not enforce them. The Guardrails screen shows each one with its current usage, the counting rule and whether it was counted in the latest scan. This screen is read only.

Guardrails (recommendations published by Atlassian)
GuardrailValue
Work items per site30,000,000
Spaces per site30,000
Space role actors per space5,000
Space roles per user per site2,000

The space role actor count is shown with and without Jira Service Management customer accounts, because native warnings have counted them. Roles per user is the highest number of roles one account holds inside a single space, a lower bound for the site total; no per-user list is stored.

Overview screen

  • Header: the time and duration of the last scan, or the time a running scan started, and Refresh now. A manual refresh is queued and allowed at most once every 15 minutes.
  • Key figures: scopes blocked now, scopes at or above 90%, scopes in the warning band and the next expected block.
  • Scores: the limits headroom score with its 90-day history, and the hygiene score (see Headroom and hygiene scores).
  • Limit cards: one card per limit with the highest usage and its scope, the band, the enforcement status, the forecast and the 90-day usage of that scope, plus a card per guardrail. Overview, Next to block, the dashboard gadget and the Rovo agent use the same calculation, so they show the same numbers. The 90 days are 90 calendar days ending today.
  • Fastest wins: up to three Safe candidates from the latest snapshot, with a link to Cleanup.
  • Migration exemption: when you entered a migration completion date, the days left and the scopes at or above 90%.
  • A notice when write mode is off, and a notice when the latest snapshot is partial, naming the steps that did not finish.

Colours: below 80% of a limit is green, from 80% up to 100% is yellow, above 100% is red. At exactly 100% LimitRadar uses yellow with the label At limit: Jira blocks adding more. These thresholds are fixed.

Next to block and forecast

Next to block lists the scopes expected to reach their limit soonest, with the limit, the scope, the usage, Why, Fastest fix and the forecast. Why names the weekly growth, a contributing scheme when the snapshot links one, or that the scope is blocked now. Fastest fix names up to five unused candidates that would bring the scope to 80%, or the number to remove.

What if I add, below the list, checks a number you enter against the top scope in the list and shows the usage after the add and whether Jira would accept it or reject the whole action. If there is no scope to check, the number is not valid or the check fails, a message under the field says so.

Forecasts use a robust linear trend (Theil-Sen) on the daily usage points of each scope:

  • 14 or more points give a forecast date (“In 24 days”).
  • 7 to 13 points give an early estimate, labelled low confidence.
  • Below 7 points there is no date, and LimitRadar shows how many points it has collected.
  • A flat or falling series shows No limit expected within 365 days. A sudden jump, for example after a scheme import, restarts the trend from the jump.
  • With at least 28 daily points, the forecast can include a day-of-week pattern (for example no new fields at weekends) when that clearly improves the fit. Forecast cards and the Limits and Next to block tables then show Weekly pattern included. Below 28 points the forecast is unchanged.

Limits screen

Pick a limit to see every scope for it, sorted by usage, with a usage bar, the band and the forecast. Fields also show the count excluding locked fields, and releases the count including archived releases. The table shows up to 50 scopes per limit. A limit that was counted with no scopes in use shows 0 of the limit, the same as Overview. Statuses per workflow show the workflow name.

Work item limits

The Work item limits screen, under Limits, lists the work items closest to the per work item limits that Atlassian documents: 5,000 comments, 10,000 worklogs, 2,000 attachments and 2,000 work item links. The scan reads a sample of work items updated in the last 30 days and keeps the top counts, with no work item text. Each user sees only work items they can browse. If Jira does not return the counts, the screen says the limits are unknown, never zero.

A Rovo action, Work item limits, is available in Advanced. It is not one of the MCP tools.

Cleanup candidates

A custom field cleanup runbook walks through each native step and its undo, then the same steps in LimitRadar.

The Cleanup screen lists candidates from the latest snapshot by type: Fields, Role actors, Grants, Workflows, Schemes and Scheme variants, up to 50 per type. Each row shows the name, who created the field when the Jira audit log still has the creation record, the safety level and the reasons. For a scheme removal, the row notes that removing a field from a scheme does not delete its values. Scheme variants show one row per scheme with the scheme name, the names of the fields involved, the safety level and the reasons.

Add candidates with the row menu, then use Simulate or save them as a plan with the plan name you type. A Do not touch row cannot be added to a plan; its menu shows why. When a type has no candidates, the screen says so and hides the plan controls.

The scan also stores release candidates (overdue, undated and stale releases), empty components, components that share a name in the same space, (migrated) field pairs and duplicate screens and workflows. They appear on the space settings page, on the Migrated fields screen and through the Rovo agent.

The Cleanup screen also lists proposal-only candidate types: Labels (near duplicates such as Billing, billing and bill-ing, or singular and plural), Filters (inactive owner, or no favourites, boards or filters using it, or a JQL reference to a custom field that no longer exists, with the missing field ids as evidence), Field options, Work types, Statuses, Priorities and Security levels. Each row shows evidence, including an approximate work item count and where the item is used. Simulate shows the effect on the limits, and the CSV export includes these rows. They cannot be added to a plan: LimitRadar stays read-only for these types and deletes nothing. Filter owners are stored as active, inactive or unknown, never by name. Near-duplicate labels still come from a sample of recently updated work items. The Labels list also includes labels Jira knows that no work item carries; those rows are proposal only (no merge, no delete, no undo) and cannot enter a plan. A filter that a dashboard gadget uses is Review, with the number of gadgets as evidence, and is never Safe.

Last used data comes from Jira with its status: tracked, not tracked or no information. LimitRadar shows the status as returned and never substitutes a default date. A field qualifies as unused after Jira has tracked it as unused for 12 months, and a release counts as stale after 24 months.

Safety levels

Every candidate has one of three safety levels, shown as their own labels so they are never confused with limit colours:

Safe
The candidate qualifies (for example tracked as unused past the age threshold) and no rule blocks it.
Review
Something needs a human decision. The reasons say what, for example a reference was found or usage data is missing.
Do not touch
A hard rule protects the item. A plan skips it.

Rules that always apply

  • Locked fields, system fields, app-managed fields, default schemes and schemes assigned to a space are Do not touch.
  • Items whose last used status is not tracked or no information are never Safe. Items first seen less than 30 days ago are never Safe unless the audit log proves an older creation date.
  • A stored reference (filter, gadget, screen, board or workflow) makes a field Review. Filters LimitRadar cannot see, or an unknown filter count, also keep a field at Review.
  • Fields carry Automation has not been checked, which keeps them at Review, until an Automation export less than 30 days old is imported on References. Options, statuses, components and (migrated) fields always carry it (see Known limitations).
  • Options referenced by a workflow or younger than 30 days, and work types or priorities still used by work items, are never Safe. Irreversible actions are never Safe. Workflows, statuses, schemes and scheme variants are always Review at most.
  • A custom rule can never make a Do not touch item Safe.

References

Each scan looks for customfield_ IDs in saved filters (JQL), dashboard gadgets, screen tabs, board configuration and workflows, and stores the references it finds. It also compares filter search totals with and without access to private filters: a gap is stored as filters that are not visible, and a refused call as unknown. Both block Safe on fields.

On the References screen, choose a candidate type and then a field or other candidate from the list, or type its ID if it is not listed. The screen shows the stored references of that item (kind, name, ID and source) and the filter visibility note. The Rovo action Field references returns the references for the field you name.

Automation import

Forge apps cannot read Jira Automation rules, so LimitRadar reads them from a file. Export your rules as JSON in Jira Automation, then choose Import automation export on References and pick the file (up to about 1.5 MB). The file is read in the app; LimitRadar does not call Automation and does not keep the file. It stores the field, status and component references it finds (rule ID, rule name and the referenced ID) on your site.

  • Fields named by an imported rule count as referenced, so they are never Safe.
  • While the export is less than 30 days old, other fields no longer carry Automation has not been checked. The export date in the file is used; when the file has none, the import time is used.
  • Safety levels change on the next snapshot. Import a newer export at least every 30 days; an older export, or none, leaves every field at Review.

Simulator and minimal set

Simulate on the Cleanup screen shows the usage before and after for every affected scope (“Payments (PAY): 683 to 681 of 700”). One field can count in many spaces, so a single removal can lower several scopes at once. The simulation writes nothing to Jira.

The Rovo action Suggest minimal set finds the smallest set of Safe candidates that brings a scope under a target percent of the limit, 80% unless you give another target.

Plans and execution

A plan is a saved, named list of cleanup items with a content hash. The Plans screen lists each plan with its status. The row menu offers Download backup (the JSON backup, limitradar-<plan id>.json), Export CSV (type, ID, name, safety, action, status, scopes, evidence, native link and instructions for every item) and Execute.

Executing a plan

  1. Simulate. Execution stays off until a simulation of the same items is stored. The Execute dialog offers Simulate plan when it is missing.
  2. Confirm. Type the plan name. If the plan has irreversible items, the dialog lists each affected object by name and asks you to type MERGE or DELETE as well.
  3. Checks. Write mode must be on, the action's own switch must be on, you must hold the Executor role if one is set, and, when two-person approval is on, the plan's current hash must be approved.
  4. Backup. LimitRadar writes the JSON backup again immediately before the run and stores its hash on each action.
  5. Run. Each item is read from Jira again first. An item that changed or can no longer be read is skipped. Writes are at least one second apart, and the run stops at the first error. One run handles up to 8 items within about 18 seconds; the rest stay planned, and you run the plan again to continue.

Writes run as the signed-in administrator, so Jira enforces your own permissions and Jira's audit log shows your account. LimitRadar never writes on a schedule.

Cleanup actions

Cleanup actions and how they are undone
ActionUndo
Move a custom field to trashRestore from trash while Jira keeps it (60 days)
Remove a field from a field schemePut the field back on that scheme. Values are not deleted; they stay hidden until the field returns.
Archive a release, or release then archiveUnarchive (and clear the released flag)
Delete an empty componentRecreate it with the stored name, space and description (the component lead is not copied)
Remove a grantRe-add the grant from the holder read before the delete
Remove a role actorRe-add the actor
Disable an optionEnable it again
Remove a work type or a priority from its schemePut it back on the scheme
Delete an option or a statusRecreated with a new ID (irreversible for the original ID)
Merge fields (fill empty targets only)No undo in LimitRadar. The source field goes to the Jira trash only when the merge finished without conflicts.
Merge releases, merge a component into anotherNo undo; work items stay on the successor
Delete an inactive workflow, or an orphaned field, workflow, permission, work type, screen or issue type screen schemeNo undo
Clone a field scheme for one space (scheme variant)No undo. The shared original scheme is not edited.

Fields always go to the Jira trash. LimitRadar has no code path that deletes a field permanently. Releases are archived and never deleted. Security levels and notification schemes are guidance only, with a link to the native screen.

Automation and other apps. Automation rules, advanced workflow features and other apps that use a field may stop working after the field is trashed. LimitRadar knows your Automation rules only from an export you import on References, so import a current export, or check your rules yourself, before you trash a field.

Actions log, undo and trash

  • Actions log: the latest executed actions with the action, the result and whether undo is still available. Each entry stores who ran it, when, the before state and the backup hash, and is chained to the previous entry's hash.
  • Undo: choose Undo on a row of the Actions log, or use the Rovo action Undo an action. A trashed field can be restored while it is in the 60-day trash window; other reversible actions stay undoable while their log entry is stored. Undo needs write mode.
  • No undo: the Actions log says No undo for every action LimitRadar cannot reverse: field merges, release and component merges, deletes of workflows and schemes, and scheme clones. The label stays even when Jira returned an ID for the deleted object.
  • Trash: every trashed field on the site with the days left before Jira deletes it, and whether LimitRadar or someone else trashed it. With write mode on, Restore on a row brings the field back, whoever trashed it. Any trashed field can also be restored in Jira's own trash.

Migrations and migrated fields

After a migration from Server or Data Center, Jira grants a temporary exemption for an entity whose usage is at or above 90% of its limit. Atlassian sources state the window differently, so you choose 3 months (support page) or 6 months (2026 announcement) in Settings. With a migration completion date, Overview shows when the exemption ends and the scopes at or above 90%.

The Migrated fields screen lists fields whose names end with “(migrated)”, paired with the original field of the same type. Each scan samples up to 24 pairs, with two pages of 50 work items each: a differing value is Conflict, equal overlap is Review, and a target that is empty on the sampled values is Safe to merge.

A merge runs from a plan, only between fields of the same type with compatible options, and only fills empty targets; conflicting values are never copied. One run reads up to four pages of work items and copies up to 40 values. If more pages remain, run the plan again. The source field is moved to trash only when the run finished its pages and found no conflict.

Field scheme variants

When a shared field scheme gives one space fields it does not show on any screen, LimitRadar proposes a variant for that space. Up to eight of those fields are checked with a JQL sample; a field with values in the sample is marked as hidden with values. Running a scheme variant plan item clones the scheme, removes those fields from the clone and associates the space with the clone. The shared original is never edited. Values in a removed field stay in Jira and are hidden until the field returns.

Duplicates and release hygiene

  • Duplicate fields: duplicate and near-duplicate field names, grouped by the Rovo action Find duplicates.
  • Duplicate configuration: screens hashed from the fields on their tabs and workflows hashed from their statuses. Matching hashes become candidates, grouped by the Rovo action Duplicate configuration.
  • Release hygiene: overdue, undated and stale releases per space.
  • Components: empty components, and components that share a name in the same space as Review merge candidates.

Space settings page

Space administrators find LimitRadar in Space settings. The page shows cards for fields, work types, security levels, releases, components and priorities with the usage for that space and a 90-day trend, and lists the local release and component candidates with their safety. Fields and work types are managed by Jira administrators, so the page explains that the schemes count for the space.

When write mode is on, a space administrator can archive a local release or delete an empty component after typing its name. The call runs as that user. A user without Administer spaces is refused on the server.

Dashboard gadget

Add the Jira limits at risk gadget to a Jira dashboard. It shows the scopes closest to a limit, each with the limit, the scope, a usage bar and the forecast or Blocked now, plus the snapshot time and a link to LimitRadar. In edit mode choose 3, 5 or 10 rows. Only Jira administrators see the data; other viewers see a short notice.

Alerts, weekly digest and scheduled reports

Weekly digest (Standard and Advanced). In both editions, the Alerts screen shows a weekly digest: one summary of the last 7 days. It lists the limits at or above the fixed thresholds (80 percent, 90 percent and Blocked at the limit, up to 10 rows) with the forecast date for 100 percent, and the biggest usage changes since 7 days ago (up to 10 rows). The weekly scheduled job builds it once a week from the stored snapshots, and the first time you open the screen it is built right away. The digest is shown in the app only; LimitRadar sends no notification for it. In Advanced, the digest stays on the Alerts screen, below the threshold alerts after each scan.

Threshold alerts after each scan Advanced. Threshold alerts after each scan are an Advanced addition to the weekly digest. Each scan raises an alert when a scope's level changes: warning from 80%, alert at 90% or when the forecast reaches the limit in less than 30 days, and blocked at the limit. A custom rule can set a different alert percent for one limit. One alert stays open per scope and level. The Alerts screen lists them, and Acknowledge on a row marks an open alert as seen. On Standard, scans raise no alerts, the alert list and Acknowledge are refused with a message, and the alert percent of a custom rule is fixed.

Alert and report space. In Settings, choose one Jira space for alerts and reports, and optionally any work type that exists in that space (the default is Task; sub-tasks are not listed, and Task is used if the chosen type was later removed). Then a weekly limits report is opened every week, and a monthly report on the first daily run of each month. In Advanced, each new threshold alert after a scan also opens a work item in that space. Alert, report and approval work items get the labels limitradar and limitradar-<kind>, plus a Kind line that Jira Automation can match. Alerts and reports are created as the app, because these jobs run without a user. Approval requests open a work item in the same space as the admin who asked. When you save a new space, LimitRadar checks that you can browse it and create work items there, and refuses the save otherwise. Until a space is chosen, nothing is created and the reports stay on the Reports screen.

LimitRadar sends no email, Slack or webhook messages.

Limits reports

The Reports screen shows the highest usage per limit with the forecast. The usage column shows the percent and the forecast column shows the forecast.

  • Standard: Export CSV, and Copy Markdown to paste the report into Confluence.
  • Advanced: Sign and export PDF. Each signed report stores a SHA-256 hash and the hash of the previous signed report, and the PDF lists your plans and the actions log. Verify recomputes the stored report's hash and compares it with the current snapshot: Verified, Mismatch, or Only the signature can be verified when the stored document or snapshot is gone. On Standard these controls show a lock.

Headroom and hygiene scores

The limits headroom score is 100 minus up to 10 points per limit. A limit scores the worse of its usage above 80% and a near forecast (10 points within 30 days, 5 within 90). Overview shows it with a 90-day history, and the tooltip shows the formula.

The hygiene score starts at 100 and drops for orphaned schemes, duplicate configuration groups and inactive workflows.

Both scores have a short explanation under them and a tooltip with the details.

Native tools

For each limit, the Native tools screen lists the native Jira screens that help with it, for example Site Optimizer for fields, field schemes, unused field options cleanup, permission schemes, space releases and components, with Open in Jira and the plan availability (all plans, Premium and Enterprise, or unknown).

Configuration (Settings)

Settings is available to Jira administrators.

Write mode
The master switch for cleanup actions, off at install, and a switch per action type. While write mode is off, the action switches are shown off and greyed out. See Write mode.
Two-person approval
Off by default. See Roles and approval.
Alert and report space
The space where weekly and monthly reports, approval requests and, in Advanced, threshold alerts after each scan open a work item of the chosen work type (Task by default), or Do not create work items. See Alerts, weekly digest and scheduled reports.
Retention
In Advanced, the number of days LimitRadar keeps history, from 30 to 2555 (400 by default). On Standard the field shows the fixed 90 days and cannot be changed. See Data, privacy and security.
Migration
Migration completion date and the exemption window (3 or 6 months).
Roles
Optional Jira groups for Viewer, Planner, Executor and Approver, chosen with a group picker. See Roles and approval.
Custom rules
Your own candidate rules, added in a dialog: a rule key, a name pattern, a scheme name pattern, a limit picked from the list of limits with an alert percent (Advanced; fixed on Standard), and a maximum number of work items with a value (checked for up to 20 fields per scan).

Settings has one Save and, right after it, one Cancel at the bottom left; nothing is stored until you save. Colour thresholds are fixed so they mean the same thing in every Radrly app. The unused field age (12 months) and the stale release age (24 months) use their defaults.

Roles and approval

Every LimitRadar screen and Rovo action requires Administer Jira. On top of that, you can pick a Jira group per role in Settings. An empty role stays open to every Jira administrator.

Planner
Creates plans. Enforced when a group is set.
Executor
Executes plans. Enforced when a group is set. Writes still run as the user.
Approver
Approves plans. Enforced when a group is set.
Viewer
Makes Jira administrators in that group read only. They can open every screen, but cannot create plans, request approval, execute, undo, restore from trash, approve, acknowledge alerts, start a scan, sign a report or import an Automation export, in the app or through Rovo. They can still change Settings. Jira administrators who are not in the group keep full access. The group does not give anyone who is not a Jira administrator access to LimitRadar.

Two-person approval is optional and off by default. When it is on, a plan runs only when its current SHA-256 content hash is approved, so any change to the plan voids the approval.

  1. Request. On Plans, choose Request approval in the plan's row menu, or use the Rovo action Request approval. When an alert and report space is set, the request opens a Task there, created as you.
  2. Decide. A different Jira administrator opens Approvals, chooses Approve or Reject on the pending row and can add a comment. The plan's author cannot approve it, and when an Approver group is set, only its members can decide.
  3. Run. Once the current hash is approved, Execute runs as usual. The Rovo action Approval status shows whether a plan is pending, approved or rejected, with the approver's name.

Write mode

Write mode is off at install. While it is off, analysis, simulations, plans and exports work, and every cleanup action is refused. A Jira administrator enables it in Settings. Each action type has its own switch. While write mode is off, these switches are shown off and greyed out, because no action can run. When you turn write mode on, each switch shows its saved state; an action with no saved switch is allowed, and an item whose action is switched off is skipped when a plan runs.

Rovo agent guide Advanced

The LimitRadar agent is part of the Advanced edition. Open Rovo chat in Jira and choose the LimitRadar agent. Rovo must be available on your site; your organisation admins manage which agents people can use, and Atlassian's Rovo usage terms apply.

Conversation starters

  • Which Jira limits will we hit next?
  • How close is space PAY to each limit today?
  • Get PAY below 80 percent fields safely.
  • Summarise our limits status for the monthly IT report.

Skills

  • limits-forecast: Ask which Jira limits you will hit next and when.
  • cleanup-plan: Ask for a safe cleanup plan, from candidates and simulation to approval, execution and undo.
  • what-if-capacity: Ask whether you can add N fields to a space.
  • field-hygiene: Ask about duplicate fields, where a field comes from or the smallest field set a space needs.

The action What if I add receives the number to add and the space, so what-if-capacity answers with the count after the add and whether Jira would accept it.

Actions

The 33 Rovo actions of the LimitRadar agent
#ActionWhat it returns or does
1Limits summaryHighest usage of each enforced limit from the latest snapshot.
2Next to blockScopes forecast to reach a limit first.
3Scope usageUsage of each limit for one space or scheme.
4ForecastForecast detail for one limit and scope.
5Will exceedScopes forecast to reach 100% before a date.
6What if I addWhether a bulk add of the number you give would be rejected in full.
7Explain a limitCounting rule, value and enforcement status for one limit.
8GuardrailsThe four guardrails with usage.
9Cleanup candidatesCleanup candidates with safety and evidence.
10Candidate evidenceEvidence for one cleanup candidate.
11Field originAuthor and creation date when the audit log still has them.
12Find duplicatesDuplicate and near-duplicate fields.
13Simulate cleanupBefore and after for the selected candidates. Writes nothing.
14Suggest minimal setSmallest Safe set that lands a scope under a target percent (80% by default).
15Create a planSaves a cleanup plan in LimitRadar. Does not change Jira.
16Plan statusPlans and item progress.
17Execute a planRuns the reversible items of a plan after confirmation, when write mode is on.
18Undo an actionRestores a reversible action while undo is available.
19Action logExecuted actions and whether undo is still available.
20Migration countdownThe migration date you entered and the 3 or 6 month window.
21Field referencesFilters, boards, gadgets, workflows and screens that reference a field.
22Health scoreLimits headroom score and hygiene score.
23Migrated fields(migrated) pairs classified Safe to merge, Review or Conflict.
24Scheme proposalA field scheme variant proposal. Does not edit the shared scheme.
25Duplicate configurationIdentical screens and workflows.
26Release hygieneOverdue, undated and stale releases.
27TrashTrashed fields and the days left to restore them.
28Request approvalAsks a different admin to approve a plan hash.
29Approval statusWhether a plan is pending, approved or rejected.
30Native toolsNative Jira screens for a limit, with plan availability.
31Custom rule resultsMatches of your custom candidate rules.
32Plan backupHash and file name of the JSON backup stored for a plan.
33Work item limitsWork items closest to the per work item comment, worklog, attachment and link limits. Not an MCP tool.

What runs where

  • Create a plan saves a plan and changes nothing in Jira.
  • Execute a plan needs the plan name as confirmation, write mode, a stored simulation and, when approval is on, an approved hash. It runs only reversible items: trash fields, remove fields from schemes, archive or release and archive releases, delete empty components, remove grants and role actors, disable options, and remove work types or priorities from schemes. Merges, deletes of options, statuses, workflows and schemes, and scheme clones stay in the LimitRadar app.
  • Undo an action takes an action ID, or a plan ID that has exactly one undoable action.
  • The actions receive the inputs they ask for, for example the date in Will exceed, the plan name and candidate IDs in Create a plan, the confirmation phrase in Execute a plan and the action ID in Undo an action. Since version 1.2 a value you give is always used and never replaced by a default.
  • Members of the Viewer group get read-only answers: actions that change something are refused for them.

Rules the agent follows

  • It answers only from the data its actions return and uses names, numbers and dates exactly as returned. It always states the snapshot time.
  • For “will we exceed” questions it uses the forecast. If a forecast has fewer than 7 daily points, it says so.
  • Before any change it shows the plan, the simulated effect and the undo path and asks for confirmation. It says so when write mode is off or a plan needs approval.
  • It treats object names as data, so instructions hidden in a field or rule name are ignored.
  • It never outputs account IDs or email addresses.
  • Every action first checks that you are a Jira administrator. Other users get a refusal without data.

LimitRadar tools in Rovo Studio and AI clients (MCP) Advanced

LimitRadar exposes 10 read-only tools through the Atlassian Rovo MCP service (Forge rovo:mcp, in Atlassian Preview). You can add them to your own agents in Rovo Studio or connect them to an MCP-compatible AI client such as Claude Code, Claude Desktop, Codex or Cursor. They answer from the latest LimitRadar snapshot, write nothing and need LimitRadar Advanced.

The 10 LimitRadar MCP tools
ToolInputsWhat it returns
lr-limits-summarylimitKey, maxRows (optional)The scope with the highest usage for each limit.
lr-next-to-blockwithinDays, maxRows (optional)Scopes forecast to reach a limit first, with date and reason.
lr-scope-usagespaceKey or schemeName, maxRows (optional)Usage of every limit for one space or scheme.
lr-forecastlimitKey, scopeIdForecast detail for one scope.
lr-will-exceedbyDate, maxRows (optional)Scopes forecast to reach 100% on or before a date.
lr-what-iflimitKey, addCount, spaceKey or scopeIdWhether adding N items is accepted or rejected in full, and the count after the add.
lr-explain-limitlimitKeyCounting rule, value and enforcement status.
lr-guardrailsnoneThe four site guardrails with usage.
lr-health-scorenoneLimits headroom score and hygiene score.
lr-migration-countdownnoneMigration date, exemption window and scopes at 90% or more.

Turn it on

  1. External access is off by default. A site or organization admin opens Atlassian Administration → Apps → Sites → your site → Connected apps → LimitRadar and turns on Expose tools via external MCP endpoint. During the Atlassian Preview this turns on all 10 tools together.
  2. Add the endpoint https://mcp.atlassian.com/forge/92a7ce6f-dd83-42f9-8c18-c199eb85de3b as a remote MCP server in your client. In Claude Code: claude mcp add --transport http limitradar https://mcp.atlassian.com/forge/92a7ce6f-dd83-42f9-8c18-c199eb85de3b.
  3. Connect. Atlassian opens an OAuth consent screen: choose your site, sign in and approve.
  4. Ask, for example: “Can we add 25 fields to space PAY?”

For Rovo Studio: create an agent, open Tools → Add tools → Connected apps, choose LimitRadar and add the tools you want.

Security and data

  • Every call runs as the signed-in user, and Atlassian does not allow app-level access for MCP calls. LimitRadar also checks that the user is a Jira administrator before it reads anything. Other users get a refusal without data.
  • Only read-only tools are exposed. Plans, execution, undo, approval requests and the Work item limits action stay in the app and the LimitRadar Rovo agent.
  • Tool results contain limit counts, scope names, forecasts and scores. They contain no work item content, field values, account IDs, names or email addresses.
  • LimitRadar still has no egress and stays Runs on Atlassian. The MCP endpoint is run by Atlassian. Your AI client handles the results under your agreement with its provider. Turn the endpoint off at any time in Atlassian Administration.

Scans and scheduled jobs

  • On install and upgrade: a scan is queued.
  • Daily: a new snapshot, the personal data report, the retention purge and, on the first day of each month, the monthly report.
  • Hourly: a new scan when a space, release or component was created, updated or deleted since the last one, or to continue a running scan.
  • Weekly: the weekly digest and the weekly report.
  • Manual: Refresh now, at most once every 15 minutes.

A scan runs in steps. Each step uses up to 12 Jira calls, saves its place and continues in the next invocation until the snapshot is finished. Jira rate limits (429 and 503) are retried up to 4 times using Retry-After.

Permissions and scopes

Every request first checks the signed-in user's Jira permissions (Administer Jira, or Administer spaces for the space page) as that user before LimitRadar reads any data as the app. Roles and space access are always decided on the server.

LimitRadar declares only granular scopes: read scopes to measure limits, scan references and read the audit log, write and delete scopes for the cleanup actions and work item writes, storage:app and report:personal-data. Version 3.0.0 removed the unused read:notification-scheme:jira scope. Version 4.0.0 adds two read-only scopes, read:dashboard.property:jira (to read which saved filter a dashboard gadget uses) and read:label:jira (to list labels Jira knows, including those no work item carries). Both are major versions, so a Jira administrator must accept the update under Manage apps. No egress or remotes were added. The full list with reasons is on the Security page.

Data, privacy and security

  • LimitRadar stores configuration snapshots, usage series, forecasts, candidates, references, plans, simulations, approvals, alerts, reports, the actions log and settings in Atlassian Forge hosted storage (Forge SQL and the Forge key-value store), following your Jira site's data residency.
  • The only personal data stored is Atlassian account IDs: field authors (from the Jira audit log), plan authors, approval requesters and approvers, action executors, people who acknowledge alerts, report signers, the person who imports an Automation export, inactive role actors and suspended grant holders that a cleanup can target, accounts checked while scanning role actors (until the next snapshot replaces them), and a user grant holder kept in the action snapshot so undo can restore the grant. Display names are resolved when a page is shown and are not stored.
  • For work item limits, LimitRadar stores work item IDs, keys, space IDs and comment, worklog, attachment and link counts for the work items closest to those limits. It does not store work item text.
  • Work item content is not stored. Field values are read only to compare sampled (migrated) field pairs, to check hidden fields for a scheme variant, to count work items for a custom rule and to copy values in a field merge that you run. LimitRadar keeps none of them.
  • A daily job reports the stored account IDs to Atlassian's personal data reporting API. When Atlassian reports an account as closed, LimitRadar replaces that account ID with a stable pseudonym everywhere it is stored and deletes the rows that held the raw ID. An account reported as updated changes nothing in storage, because display names are read from Jira when a page is shown.
  • A daily job deletes data older than the retention period: 90 days on Standard, and the retention setting in Advanced (30 to 2555 days, 400 by default). It covers daily usage and headroom score history, action log entries, alerts, plan backups, plans and finished snapshots. The newest finished snapshot is always kept, and signed reports are kept until uninstall.
  • After a downgrade from Advanced to Standard, the stored retention setting is kept and the effective retention becomes 90 days. Nothing is deleted at once: the normal daily job removes the older history in batches. If you upgrade again, the stored setting applies to the data that is still there.
  • No data leaves Atlassian. See Privacy, Security and the DPA.

Editions

LimitRadar comes in two editions, licensed through Atlassian Marketplace.

  • Standard: all limits and guardrails, forecasts, Next to block and What if I add, cleanup candidates, the simulator, plans with backup and CSV, all cleanup actions with undo wherever Jira allows it, field authors from the audit log, the reference scan, migrated field merges, the headroom and hygiene scores, custom rules with fixed alert thresholds, roles, write mode, field scheme variants, the trash view, the space settings page, the dashboard gadget, the weekly digest in the app, native tools, the Work item limits screen, cleanup proposals with evidence (labels, filters, field options, work types, statuses, priorities and security levels), the limits CSV and Markdown, history kept for a fixed 90 days, and the interface in 12 languages (English, German, Japanese, French, Spanish, Brazilian Portuguese, Polish, Korean, Italian, Dutch, Simplified Chinese, Traditional Chinese).
  • Advanced: everything in Standard, plus the Rovo agent (all 33 actions) and its 10 read-only MCP tools for Rovo Studio and AI clients, threshold alerts after each scan with configurable alert thresholds, configurable retention (30 to 2555 days) and the signed PDF limits report with the hash chain and Verify.

If the licence lapses, LimitRadar stays read only for 30 days while snapshots continue: reads work, and cleanup actions, the Rovo agent and the signed PDF are refused. After 30 days the app is blocked.

On Standard, every Rovo action answers that the agent is part of Advanced and returns no data. Alert lists, acknowledging alerts, changing an alert percent and changing retention are refused with a message, and the screens explain what Advanced adds.

Languages

The interface, server messages and Rovo notes follow the language in the user's Jira profile, in 12 languages (English, German, Japanese, French, Spanish, Brazilian Portuguese, Polish, Korean, Italian, Dutch, Simplified Chinese, Traditional Chinese), with English as the fallback. Settings has no language picker. Object names (fields, spaces, schemes) are shown as they are in Jira and are not translated.

Known limitations

These are known edges of the current release.

Automation is checked from a file you import
Forge apps cannot read Jira Automation rules. LimitRadar uses the JSON export you import on References, applies it on the next snapshot, and treats it as current for 30 days. Without a current export, fields stay at Review. Option, status, component and (migrated) field candidates always carry Automation has not been checked and stay at Review, never Safe; you can still plan and run Review items after checking Automation yourself.
Alert and report work items use the chosen work type
Alerts, the weekly and monthly reports and approval requests open a work item in the alert and report space, using the work type chosen in Settings (Task by default). If the chosen type was removed, Task is used. If that type cannot be created, Jira refuses the create; the scan still finishes and a threshold alert stays on the Alerts screen.
The Viewer group only restricts Jira administrators
LimitRadar is for Jira administrators. The Viewer group makes its members read only; it does not open LimitRadar to anyone else. Members can still change Settings. If Jira does not answer the group lookup, LimitRadar does not lock the administrator out.
Release and component merges cannot be undone
Jira does not offer an unmerge. After LimitRadar merges a release into another release, or a component into another component, the work items stay on the successor. The dialog asks you to type MERGE as well as the plan name before it runs.
Deletes of workflows and schemes, scheme clones and field merges have no undo
These actions are irreversible in LimitRadar, and the Actions log says No undo for them. Download the plan backup before you run them.
A recreated option or status gets a new ID
Undo creates a deleted option or status again, and Jira assigns a new ID. LimitRadar does not write that new ID back into workflows that referenced the old one. Check those workflows yourself first.
Field authors come from the audit log only
When the Jira audit log still has the field creation record, Cleanup shows who created the field. There is no field-created trigger, so a field whose audit record has rolled off has no author in LimitRadar.
Migrated fields are compared from a sample
Each scan samples up to 24 pairs, with two pages of 50 work items for each pair. A pair that was not sampled, and any work item outside those pages, is shown as Review. A merge copies up to 40 work items per run; run the plan again to continue.
Some cleanup types are proposals only
Labels, filters, field options, work types, statuses, priorities and security levels appear as cleanup proposals with evidence. They cannot be added to a plan, and LimitRadar does not delete them. Security level deletion stays guidance only, with a link to the native screen.
A long scan continues in steps
A scan step stops after 12 Jira calls and saves its place, and the next invocation resumes the same snapshot. A step that fails is marked on Overview, and LimitRadar does not invent the count.
Fixed settings
The unused field age (12 months) and the stale release age (24 months) have no control in Settings in this release.

Roadmap

The items below have shipped in LimitRadar 3.1.0, 4.0.0, 4.1.0 and 4.2.0. There is nothing Planned on this roadmap.

Shipped

  • Fields created by an app account stay at Review. A custom field whose creator in the Jira audit log is an app account is capped at Review and is never Safe. Locked, system and app-managed fields stay Do not touch. Both editions. Version 4.2.0
  • Review request. After the first completed scan, Jira admins see one message at the top of Overview, Enjoying LimitRadar?, with Write a review, I already reviewed and Not now. Write a review opens the Marketplace reviews page in a new tab, with no tracking. The answer is stored once per site with no personal data. Both editions, all 12 languages (English, German, Japanese, French, Spanish, Brazilian Portuguese, Polish, Korean, Italian, Dutch, Simplified Chinese, Traditional Chinese). Version 4.1.0
  • Filter hygiene for deleted fields. Cleanup > Filters flags saved filters whose JQL names custom fields that no longer exist, and lists the missing field ids as evidence. It is review only and never deletes or writes the filters. Both editions. Version 3.1.0
  • Findings as Jira work items. Admins can pick any work type that exists in the alert space. Alert and approval work items get the labels limitradar and limitradar-<kind>, plus a Kind line that Jira Automation can match. Threshold alerts after each scan are Advanced. Version 3.1.0
  • Dashboard gadget filters. The scan reads which saved filter a dashboard gadget uses. A filter that a gadget uses is Review, with the number of gadgets as evidence, and is never Safe. This is read only. Both editions. Version 4.0.0
  • Unused label inventory. Cleanup > Labels also lists labels Jira knows that no work item carries. It is read only and proposal only: no merge, no delete, no undo, and the rows cannot enter a plan. Both editions. Version 4.0.0

A shorter overview is on the home page roadmap.

FAQ

Do the limits depend on my Jira plan?

No. Atlassian publishes the same limits for every plan. The plans differ in tooling: Site Optimizer is available on Premium and Enterprise. LimitRadar works on every plan and adds forecasts and guarded cleanup with undo. The Rovo agent, threshold alerts after each scan, configurable retention and the signed PDF report are Advanced.

Will LimitRadar change anything in Jira on its own?

No. Analysis is read only, write mode is off at install, and every write is an explicit command by an administrator in the app or a confirmed Rovo action. There are no scheduled writes.

Can LimitRadar delete a field permanently?

No. Fields always go to the Jira trash, where they can be restored for 60 days. Trashed fields do not count toward the fields limit.

How does LimitRadar know whether a field is used?

It combines Jira's last used data (with its status), the references found in filters, gadgets, screen tabs, boards and workflows, and the visibility of private filters. Items without reliable usage data are never marked Safe.

Does LimitRadar check Automation rules?

Through an export you import. Forge apps cannot read Automation rules, so export them as JSON in Jira Automation and import the file on References. Fields named by a rule are treated as referenced, and while the export is less than 30 days old the other fields are no longer held at Review for Automation. Without a current export, field candidates stay at Review.

How is LimitRadar related to ConfigRadar?

They are separate apps from the same family. ConfigRadar records what changed in your configuration and who changed it. LimitRadar measures how close you are to Jira's limits and helps you clean up.

What happens when Atlassian changes a limit?

The limit catalog ships inside the app. A change arrives as a new LimitRadar version. Usage history keeps the limit value of each day.

What happens if my licence expires?

For 30 days after the licence lapses, LimitRadar stays read only and snapshots continue: you can read results, and cleanup actions, the Rovo agent and the signed PDF report are disabled. After 30 days the app is blocked. Renew the licence through Atlassian Marketplace to continue.

Troubleshooting

A limit card says “Not available: missing permission” or “Not counted: the scan step failed”
LimitRadar could not read one of the endpoints for that limit. The rest of the app keeps working. Check that the app's scopes were approved, use Refresh now, and contact support if it persists.
My numbers differ from what Jira shows
Check the last scan time in the header and use Refresh now. For fields, compare the count with the count excluding locked fields. For releases, compare the counts with and without archived releases.
A forecast says it is collecting trend data
Forecasts need at least 7 daily points for an estimate and 14 for a date. The label shows how many points exist.
Plan items were skipped
An item is skipped when it changed or could not be read since the plan was saved, when its action is switched off, or when it is Do not touch.
Execute does nothing or shows an error
Write mode is off, the plan has no stored simulation, the typed plan name or confirmation word does not match, you are not in the Executor group, you are in the Viewer group, or two-person approval is on and the plan's current hash is not approved (see Roles and approval).
Only part of a plan ran
One run handles up to 8 items within about 18 seconds and stops at the first error. Run the plan again to continue.
The Rovo agent refuses to answer
The agent is Advanced and works for Jira administrators only. Check that you are on Advanced, that Rovo is enabled for you and that the LimitRadar agent is allowed in your organisation.
Undo is not offered
The Actions log says No undo because the action is irreversible, the 60-day trash window of a field has passed, write mode is off, or you are in the Viewer group.

Still stuck? See Support.

More Radrly apps for Jira