Radrly
LeakRadar
Try it free

Documentation

Everything you need to install, configure and use LeakRadar.

Documentation for LeakRadar 2.32.0, updated 26 October 2026. It describes the Forge production version built from the main branch at commit 375233e (engine 0.2.0).

Overview

LeakRadar is an Atlassian Forge app that scans Jira Cloud and Confluence Cloud for credentials and personal data. Deterministic detectors find candidates, Forge LLMs review the ambiguous ones, and your team triages and remediates the findings. LeakRadar never stores the full value it found.

Product principles:

  • LeakRadar never stores and never shows a full secret or personal data value.
  • Every finding has a band, a severity and a reason. A false positive can become a suppression so it does not come back.
  • Remediation starts with rotating the secret. Deleting text comes second, because Jira history and Confluence page versions keep old content.
  • Zero egress. The AI runs inside Atlassian and sees only masked context, with every detected value replaced by a placeholder. AI triage is on by default and a Jira administrator can switch it off.

Requirements

  • Jira Cloud (required). LeakRadar is installed on Jira and its licence follows your Jira users.
  • Confluence Cloud on the same site (optional). Connect Confluence to LeakRadar to scan it with the same licence.
  • A Jira administrator for installation, settings and the full admin page.
  • Rovo only for the Rovo agent.

Install and get started

  1. Install LeakRadar on your Jira site from its Atlassian Marketplace listing and approve the requested scopes. To scan Confluence, connect LeakRadar to Confluence in Atlassian Administration; pages and comments are scanned only after that connection exists.
  2. Open Get started (Jira settings → Apps → Get started with LeakRadar). It walks through the install, the Confluence connection and the first scan, and shows the scan state.
  3. Choose the scope on Scan → Overview: pick Jira projects, Confluence spaces and work item types (empty means all), whether to include personal spaces and archived projects and spaces (both off by default), attachments, text in images and history (all on by default).
  4. Start the first scan. The backfill covers the last 24 months by default, or all time, newest content first. Product events keep scans live afterwards.
  5. Add your team. Add people to the LeakRadar team on Scan → Overview so they can work on the Findings page, or grant delegated roles on Access → Exposure for one project or space.
  6. Triage critical findings first. Open Findings, filter by severity and status, rotate the secret and clean up the content.

Where LeakRadar appears

Admin page (Jira settings → Apps → LeakRadar)
Sections Scan (Overview), Detection (Built-in detectors, Custom detectors), Audit (Evidence, Erasure) and Access (Visibility, Exposure, Policies). LeakRadar configuration opens the same page on the detectors.
Get started (Jira settings → Apps)
The install steps and the start of the first scan.
Findings (Jira global page, Apps → LeakRadar)
The findings list and the finding detail, for Jira administrators, members of the LeakRadar team and people with a delegated role.
Project settings → LeakRadar
Findings of that Jira project for its project administrators and people with a role there. Jira administrators and people with the delegated Admin role also set the project's default sensitivity label there (Container label).
Space settings → LeakRadar (Confluence)
Findings of that space for its space administrators and people with a role there, with the same Container label default.
Confluence administration → LeakRadar
Confluence findings and a link to the Jira admin page, for Jira administrators.
Work item panel and page byline
A LeakRadar panel on Jira work items with a lozenge such as 2 open or Clear, and a byline item on Confluence pages, both with masked details of findings you can see. An administrator can hide both with Show the byline and glance on the Overview.
Open leaks dashboard gadget
Counts for Jira projects you administer.
Rovo
The LeakRadar agent in Jira and Confluence.

Scan scope

The scope on Scan → Overview decides what LeakRadar reads. Search and pick Jira projects, Confluence spaces and work item types; an empty list means all. Personal spaces and archived projects and spaces stay out until you turn them on. The backfill, live product events and the hourly reconcile all use the scope saved with the single Save button.

When a new scope leaves out projects or spaces that still have open findings, LeakRadar shows how many and asks what to do with them: Keep them, Hide (they leave the findings lists and are deleted after the retention period) or Resolve. Each hidden or resolved finding records the change in its decision history, and the audit log records the count. Findings do not store the work item type, so a narrower type list applies to new scans only.

Backfill and incremental scans

  • Backfill: queued jobs per Jira project and Confluence space, resumable after failure, with progress per product (items, findings and errors, which are codes, not content). Pause and resume the scan, or cancel the backfill after a confirmation.
  • Incremental scans: LeakRadar follows Jira events for created, updated and deleted work items, comments and attachments, and Confluence events for pages, blog posts, comments and attachments, then fetches the current content to scan it.
  • Reconciliation: an hourly job queues new projects and spaces and re-scans work items updated in the last two hours, which catches missed events.
  • Re-scan on demand: redaction and attachment actions queue a re-scan of the item, and the Rovo agent can queue one.
  • Full re-scans: weekly, monthly or off, per product, set on the Overview. A detector engine update queues one full pass.
  • App upgrades: when a new LeakRadar version is installed while a scan runs, the scan pauses and the next hourly reconcile resumes it.
  • Coverage: items recorded and items in the current backfill, per product.

What is scanned

Content LeakRadar scans
ProductContent
ConfluencePages and blog posts (title and body), footer and inline comments, up to 15 older versions of a page or blog post, attachments
JiraSummary, description and comments; field history of summary, description and environment (the latest 20 changelog entries); attachments
AttachmentsThe first 10 files of an item, up to 5 MB each: text formats, the PDF text layer, docx, xlsx and pptx text, HAR files. With OCR: PNG, JPEG, GIF and WebP images and images inside PDFs without a text layer

OCR runs inside the scan function with a bundled English Tesseract model; nothing is downloaded. Images under 40 pixels on a side or over 3 megapixels are skipped. Not scanned: Jira custom fields other than those above, video and audio, whiteboards, databases and live docs.

Detectors

Engine 0.2.0 has 63 built-in detectors using prefixes, validators, checksums, entropy and keyword context. Every finding records the engine version.

  • Credentials (33): AWS access and secret keys, Google Cloud API keys, service account keys and OAuth client secrets, Azure storage keys and Azure AD client secrets, GitHub, GitLab and Atlassian tokens, npm, PyPI and Docker tokens, Slack tokens, Stripe keys and webhook secrets, Twilio, SendGrid, Mailgun, Square, Shopify, DigitalOcean, HashiCorp Vault, New Relic, Grafana, Pulumi, Databricks and Postman tokens, OpenAI and Anthropic keys, JSON Web Tokens and password assignments.
  • Private keys and connection strings: PEM private key blocks; database and broker URLs with credentials.
  • Payment and financial: card numbers with Luhn and prefix checks, IBAN with mod-97, SWIFT/BIC.
  • National IDs (19): PESEL, NIP, REGON and the Polish ID card, Steuer-ID, NIR, DNI and NIE, codice fiscale, BSN, personnummer, NINO, T.C. Kimlik, SSN, SIN, CPF, Aadhaar, PAN and ICAO passport MRZ lines, with checksums where they exist.
  • Health: NPI, DEA numbers and medical record numbers near keywords.
  • Contact data: email addresses (off until you enable them) and phone numbers.
  • Code integrity: Trojan Source bidirectional control characters in code blocks.
  • HAR files: authorization, cookie and proxy-authorization headers and secret query parameters.

Turn detectors on or off on Detection → Built-in detectors, or turn on a framework pack (GDPR, PCI DSS, HIPAA, SOC 2) in one step. Known sample values from documentation are discarded.

Custom detectors

On Detection → Custom detectors, add your own detector with a name, a regular expression, optional keywords, an optional validator, a severity and a sample text to test it. Patterns are checked so that a bad expression cannot slow scans down. A custom detector can be marked auto-safe for policy redaction. The Standard edition allows 20 custom detectors.

AI triage

Each candidate gets a band: definite, ambiguous or discard. Only ambiguous candidates, at most 20 per item, go to Forge LLMs, the Atlassian-hosted models available to Forge apps.

  • What the AI sees: the detector, severity and band, and up to 120 characters of text on each side of the candidate, with the whole context capped at 2,000 characters. The candidate and every other detected value in that text are replaced with a placeholder that gives only type, length and character set, never a prefix or any character of the value.
  • What it returns: a label, a confidence and a reason of up to 200 characters, shown in the finding detail.
  • Authority: the AI labels ambiguous candidates only. A definite value near a candidate appears only as a placeholder, and the AI never resolves or hides a definite finding.
  • Prompt injection: content is treated as untrusted data, the model has no tools and its output can only set the fields above.
  • Control: AI triage is on by default. A Jira administrator switches it off on Built-in detectors (Verify ambiguous findings with Forge LLM). When it is off, ambiguous candidates stay visible without an AI label.
  • Model tier: Haiku by default, or the larger Sonnet. If Sonnet is not available to Forge, LeakRadar uses Haiku.
  • Monthly budget: the number of ambiguous candidates the AI may review per UTC month, 500 by default (0 to 100,000; 0 stops AI review). Above the budget, candidates stay visible and unclassified. The meter on Built-in detectors shows this month's usage after the first triage.

Prompts and responses are not stored. LeakRadar keeps monthly usage counters.

Detection methodology

Built-in detectors are pattern and checksum checks in nine families: credentials (33), national IDs (19), health (3), financial (2), contact data (2, email off until you enable it), private keys (1), connection strings (1), payment cards with a Luhn check (1) and code integrity (1). Custom detectors are your own rules and are not part of these families.

A hit is definite when a format check such as a checksum or a known token prefix confirms it, or ambiguous otherwise. Ambiguous hits go to Forge LLMs only while AI triage is on and the monthly budget allows; the prompt carries masked context in which every detected value is a type, length and character set placeholder. Hits over the budget stay unclassified and are not deleted.

  • Benchmark: the release gate is a synthetic corpus of 10,000 documents with 500 injected values and 500 look-alike traps, with a target of at least 90 percent precision and 95 percent recall on known formats. That corpus has not been scored for engine 0.2.0, so LeakRadar publishes no precision or recall figures yet.
  • Continuous integration: a short labelled sample runs through the same gate on every build. It is not the corpus result and does not call Forge LLM. The 80-prompt Rovo check covers the answer policy (no raw value in an answer), not detection quality.
  • What a clean scan does not prove: content outside the scope, detectors you turned off, names, organisations, postal addresses and dates of birth, Jira field history that cannot be purged and images over the monthly OCR limit are not covered, and LeakRadar does not check whether a matched secret still works.

Masking and hashing

Masked snippet
Provider tokens of 20 or more characters keep a four-character prefix and the last four characters, so the owner knows which key to rotate. Connection strings keep the prefix only. National IDs and other personal data are fully masked; email addresses are reduced to a pattern such as j•••@•••.com.
Keyed hash
An HMAC-SHA256 of the normalised value with a per-installation secret key in Forge secret storage, used to group the same value across locations and for DSAR lookups. Plain hashes are not used because short values could be guessed.
Location
Product, project or space, item, and field, comment, version or attachment.

The full value is held in memory only during the scan. It is never written to storage or logs. Triage notes and suppression reasons pass through every detector before they are stored, so a pasted token is masked.

Severity and exposure

Each detector has a default severity (critical, high, medium or low), and a triager can change it. A public container raises severity by one step.

The exposure score (0 to 100) adds anonymous access (30), site-wide access (20), extra locations of the same value (up to 15), age (up to 15), severity (up to 20) and history-only presence (10). Group member counts and guest lists are not read. The access explorer on Access → Exposure shows the stored exposure of a finding, the Jira security level and the Confluence read restrictions of an item the caller can read.

Findings

The Findings list filters by product, severity, status and project or space, shows overdue findings only on request, saves views per person and selects findings for bulk triage. Each row shows the masked value, where it is and how many places share the value. An inbox line lists critical and high items, overdue findings and scans with errors; for members of the LeakRadar team it lists only items they can see and leaves out scan errors.

The finding detail shows the masked value, where it is, the location, the band, the proposed sensitivity label, the AI label and reason, status, severity, owner, due date, reason and the history of decisions, with the actions described below.

A finding is shown only to users who can see the source item. Others see a count such as “1 more findings are in content you cannot see”, without title, snippet or link.

Triage workflow

  • Statuses: Open, Confirmed, Resolved, False positive and Accepted risk. False positive and accepted risk need a reason. Closed statuses can be reopened.
  • Owner: picked by name; the account ID is stored, not the display name.
  • Due date and SLA: due dates default to 2, 7, 30 and 90 days for critical, high, medium and low, and you can set your own. Overdue findings can be filtered.
  • Bulk triage: up to 50 visible findings at once.
  • Audit trail: every decision records actor, time and the before and after state, without the value.

Value groups

The same value in many places is grouped by its keyed hash, for example “3 places”. Rotate it once and clean it up everywhere. A remediation work item is shared by the whole group.

Suppressions

Suppress a value, a detector in a space, or a location, with a reason, for 1 to 365 days (the finding detail offers 90 days). An active suppression blocks a new or reopened finding. Active suppressions are listed on the Findings page and can be undone.

Remediation playbooks

  1. Rotate or revoke the secret at its provider.
  2. Redact the value in the current content.
  3. Purge old Confluence versions that still contain it.
  4. Check Jira history and restrict the work item if needed.

Rotate anyway. LeakRadar cannot check whether a secret is still active, because that would need a call outside Atlassian. Treat every confirmed secret as live.

Assisted redaction

From Redact in place, replace the value with [REDACTED by LeakRadar] in the current version of a Jira summary, description or comment, or a Confluence page, blog post or comment. The dialog shows the masked value, the location and the rotate-first playbook. The edit runs as you, after you confirm, so your permissions apply and the product history shows your account. Formatting stays; a value that crosses formatting tags is left unchanged. The finding is then resolved and the item is queued for a re-scan.

Version purge and history

  • Confluence: after a finding is confirmed, Purge old versions lists older page or blog post versions whose body still matches the keyed hash and deletes the ones you choose, as you, in a second step. It is available to a Jira administrator who administers that space. The current version is never deleted.
  • Jira field history: apps cannot remove Jira field history. A value that was edited out becomes a history finding; rotate the secret and restrict the work item. Confirmed history findings can be handed to Atlassian Guard with the Guard handoff CSV (see Evidence packs).
  • Version grouping: the same value in consecutive versions is one finding.

Attachments

Attachment findings work like other findings. After a finding is confirmed you can Delete attachment, or Replace with a sanitised HAR: LeakRadar uploads a copy without the secret headers and deletes the original only when the upload succeeded. Both run as you after a second confirmation.

Restrict access, labels and marker

  • Restrict access: sets the Jira security level you chose on Visibility, or a Confluence read restriction, as you. The restriction can be undone.
  • Sensitivity labels: LeakRadar's own labels Public, Internal, Confidential and Restricted (content labels lr-public to lr-restricted), with names you choose, a default per project or space, optional inheritance to up to 25 child pages, and up to 25 items per apply. A higher label is not lowered without a reason. They are separate from Atlassian data classification.
  • Warning marker: optionally marks items that still have open findings with a label and a property (Jira JQL can use issue.property[leakradar].open). Off by default.

Work items and notifications

  • Remediation work items: a Task in the Jira project you choose, one per value, with the detector, severity, location, masked snippet and the rotate-first playbook, never the value. The rule for new critical and high findings is off until you turn it on; Open remediation work item does the same for one finding as you. Closing the work item does not close the finding.
  • Notifications: the owner can be notified in Jira; the message omits the secret. The in-app inbox line on Findings lists critical, high and overdue items.
  • Author notice: optional and off by default (Tell the content author on the Overview). When a scan stores a new finding in the current content of a work item, page or blog post, the app adds a comment: LeakRadar found something to review on this item. Open LeakRadar for the masked finding. This message does not include the value. No notice is posted while the licence is inactive.
  • Real-time critical alerts (Included in Advanced): each new critical finding appears at once in an alert box on Findings, with the detector, the location and the severity, never the value. Mark all as seen clears it. Optionally the LeakRadar team is also notified on the affected Jira work item through a Jira notification sent by the app. Turn both on or off under Critical alerts in Settings. Alerts are kept 90 days.
  • Weekly digest (Standard and Advanced): every Monday (UTC) LeakRadar summarises the new findings of the last 7 days on Findings: counts by severity, top detectors and top locations, without values.
  • Slack, Teams, email: LeakRadar sends nothing outside Atlassian. Use your own Jira Automation rules on LeakRadar work items if you want alerts there.

Auto-remediation policies

Policies on Access → Policies are opt-in rules that act on definite findings: open a work item, apply a label, restrict access, or redact. Redaction applies only to critical credentials and to custom detectors marked auto-safe. A new policy runs as a dry run for 7 days and records what it would have done. After that an administrator can switch it live; a live run writes as the administrator who signed the policy, through Forge offline user impersonation. After a policy redaction LeakRadar adds a comment to the item that tells people to rotate the key, without the value. Every run is listed under Recent runs.

DSAR and erasure

On Audit → Erasure, enter an email address, phone number or national ID. LeakRadar hashes it with the installation key and lists findings that already have that hash. The typed value is not stored or written to the audit log; the lookup is kept for 15 minutes as hashes only. Only values that LeakRadar has already found can match; resolved and false positive findings are left out.

Erasure redacts the current text as the confirming administrator, can purge older Confluence versions (off by default) and delete attachments, and lists Jira history locations that apps cannot delete. The erasure record stores counts (redacted, purged, history and failed) and a SHA-256 hash, which is also the next link in the evidence chain. The Rovo agent never accepts the value and points to this screen.

Evidence packs

  • Edition: signed evidence packs are included in Advanced.
  • Packs: created on demand on Audit → Evidence, and on the schedule set under Recurring compliance pack in Settings: weekly (Monday, covering the previous Monday to Monday, UTC), monthly (the first day of each month, UTC, the default) or off.
  • Compliance brief: each recurring pack adds a brief to the JSON and the PDF: the audit summary with the report hash, the brief for the period, a period comparison of created and resolved findings against the previous period with deltas, and the status of each policy. It holds no values. Recurring packs are listed under Audit → Evidence with download and Verify.
  • Files: a findings CSV (masked, up to 200 open critical and high findings), a decisions CSV, a JSON document and a PDF. The decisions CSV also lists settings changes, scan starts, pauses, resumes and cancels, re-scan cadence changes and scope narrowing, as setting names and values without account IDs or free text.
  • PDF and JSON contents: period, coverage and limitations, methodology, items scanned, open findings by severity and category, median hours to resolve, SLA breaches, the retention period, a control mapping for ISO/IEC 27001:2022, SOC 2 and GDPR (evidence for your auditor, not a certification), open critical findings with masked snippets, and the CSV, JSON and previous report hashes.
  • Chain and signature: each pack links to the previous pack's JSON hash and is signed with an HMAC-SHA256 keyed by the installation key. The report chain shows whether each signature matches.
  • Verify a file: the hash is computed in your browser; a file that is not in the chain does not match.
  • Monthly attachment: if a remediation project is set, the recurring PDF is attached to one LeakRadar evidence pack work item there.
  • Retention: packs and their files are kept for 7 years.
  • Guard handoff CSV: on Audit → Evidence, a Jira administrator downloads up to 200 confirmed Jira history findings with the work item ARI, the field, the location and the keyed hash, never the value, for bulk redaction in Atlassian Guard. It stays available while the licence is inactive, and every download is recorded in the audit log.

Work item panel, byline and gadget

The LeakRadar panel on a Jira work item and the byline item on a Confluence page list open findings you can see, with severity, detector and masked value. The Open leaks dashboard gadget shows critical and high counts, SLA breaches, the number of projects, the oldest open finding and four weeks of new findings, only for Jira projects the viewer administers (up to 50 projects). With Show the byline and glance off, the panel and the byline show no findings; scans continue.

The project settings and space settings pages show the findings of that project or space to its administrators and to people with a delegated role there. The key comes from the page context, never from the browser.

Configuration

Scan → Overview
Scope pickers, the LeakRadar team, critical alerts, the recurring compliance pack schedule, Show the byline and glance, Tell the content author, remediation work items (project and the automatic rule), attachments, text in images, Jira and Confluence history, full re-scan cadence per product, coverage, pause, resume and cancel.
Detection → Built-in detectors
Retention of closed findings (default 395 days; 90 to 731 days on Standard, up to 2,557 days on Advanced), AI triage on or off, model tier, monthly candidate budget and usage meter, framework packs and detector toggles.
Detection → Custom detectors
Your own detectors with a test box.
Audit → Evidence and Erasure
Evidence packs, Verify, the Guard handoff CSV, DSAR lookup and erasure records.
Access → Visibility
Warning marker, marker label, Jira security level for Restrict access, label names, child page inheritance and the project or space default label.
Access → Exposure
Access explorer and delegated roles.
Access → Policies
Auto-remediation policies and their runs.
Project settings and space settings → LeakRadar
The default sensitivity label for that project or space.

Scans, settings changes, scope narrowing, triage, suppressions, labels, restrictions, redaction, version purges, work items, evidence and Guard exports, erasure, role grants and closed-account pseudonymisation are recorded in the audit log, without values.

Access and delegated roles

Jira administrators (Administer Jira) see every screen. On Access → Exposure they grant delegated roles to people for one Jira project or Confluence space:

Viewer
Reads findings there.
Triager
Changes status, owner and due dates.
Remediator
Also redacts, restricts and labels in that project or space.
Admin
Full LeakRadar access for that project or space, including its default sensitivity label.

LeakRadar team: on Scan → Overview a Jira administrator adds up to 50 people by name; only their account IDs are stored. Team members use the Findings page as administrators do: triage, bulk triage, suppressions, saved views, remediation work items and attachment actions, always limited to items they can see and run as themselves. Redaction, restrictions and labels need a delegated Remediator role or Jira administration; version purge and settings stay with Jira administrators.

Project administrators and space administrators also see their own LeakRadar page. Every action also requires that you can see the source item, and every interactive edit runs as you, so Jira and Confluence apply your own permissions.

Licence and Standard limits

  • Inactive licence: LeakRadar becomes read-only at once. Findings, evidence, the Guard handoff CSV, DSAR lookups and erasure records stay available. Actions that change data are rejected on the server, and their buttons are disabled with a tooltip that explains why.
  • Grace period: scans continue for 7 days after the licence becomes inactive, then pause until it is active again.
  • Standard limits: 20 custom detectors, OCR for 2,000 images per UTC month (other files are still scanned), and retention of closed findings from 90 days to 24 months.
  • Advanced: up to 100 custom detectors, OCR with no monthly cap and retention up to 7 years. See Editions.
  • Moving from Advanced to Standard: nothing is deleted at once. A retention period above 24 months is kept for 90 days from the first daily run on Standard, and the setting is read-only during that time. After 90 days the daily job uses 731 days, so closed findings older than that are deleted. Moving back to Advanced, or saving 731 days or less, ends the grace period.

Rovo agent guide

The Rovo agent is included in Advanced. LeakRadar adds an agent named LeakRadar to Rovo, in Jira and in Confluence. Rovo must be available on your site. The agent has 34 actions: 22 read and 12 that change data or open the app.

Read actions
ActionWhat it does
lr-list-findingsList findings you can see, with filters
lr-get-findingShow one finding
lr-explain-findingExplain why a finding matched and what to do
lr-whereWhere a type of data appears
lr-summaryOpen findings by severity and product
lr-top-areasProjects and spaces with most findings
lr-value-groupAll places that share one value
lr-remediation-planOrdered plan: rotate, redact, purge, restrict
lr-sla-statusOverdue and due-soon findings
lr-compare-periodsCompare two periods
lr-audit-summaryAudit summary with the latest report hash
lr-coverageWhat was scanned and skipped
lr-detector-infoWhat a detector finds
lr-dsar-lookupPoints to the Erasure screen; refuses a pasted value
lr-page-checkIs this page safe to share
lr-who-can-seeExposure and restrictions of an item you can read
lr-exposure-topItems with the highest exposure
lr-classify-suggestProposed sensitivity label
lr-policy-statusPolicies, dry runs and recent runs
lr-erasure-statusErasure records
lr-weekly-briefShort weekly brief
lr-ai-exposure-adviceAdvice on sharing content with AI tools
Actions that change data or open the app (each asks for confirmation)
ActionWhat it does
lr-set-statusSet status
lr-assign-ownerAssign an owner
lr-set-due-dateSet a due date
lr-suppressSuppress a value, detector in a space, or location
lr-create-work-itemsOpen remediation work items
lr-rescanQueue a re-scan
lr-prepare-redactionLink to the redaction dialog in the app
lr-apply-labelApply a sensitivity label
lr-restrict-itemRestrict an item
lr-create-exceptionRecord an accepted risk with a reason
lr-schedule-rescanSet the full re-scan cadence
lr-start-erasurePoints to the Erasure screen; refuses a pasted value

Rules the agent follows

  • It answers only from LeakRadar results and cites finding IDs. It says a page is safe to share only when that item was scanned and has no open findings.
  • It never outputs, reconstructs, guesses or asks for a secret or personal data value. Results are scrubbed before they return.
  • It never claims a secret is inactive or rotated.
  • Before any change it states exactly what will change and waits for confirmation. Every action is recorded in the audit log with the channel rovo.
  • Policy go-live, redaction and erasure stay in the app.

Skills

  • triage-findings: Ask it to triage your open critical findings and set status, owner and due date.
  • remediation-plan: Ask for a remediation plan for a space or area.
  • exposure-check: Ask whether a page is safe to share and who can see it.
  • compliance-brief: Ask for the audit summary, the weekly brief or a DSAR lookup.

No skill ever reveals a secret or personal data value, and every change waits for your confirmation.

Permissions and scopes

Before any read as the app, LeakRadar checks on the server that you are a Jira administrator, a member of the LeakRadar team (for the Findings page), an administrator of that project or space, or hold a delegated role there, and that you can see the source item. Interactive writes run as you after a confirmation. Live auto-remediation policies run as the administrator who signed them. The monthly evidence work item and the optional author notice are created by the app and contain no values. LeakRadar requests 17 scopes; the full list with the reason for each is on the Security page.

Data, privacy and security

  • Stored: masked snippets and masked titles, keyed hashes and locations, AI labels and reasons, statuses, owners, due dates and decisions, triage notes and suppression reasons (masked before storage), saved views, labels with restriction undo data, policies and their runs, custom detectors, role grants, the LeakRadar team list, settings, evidence packs and erasure records, the audit log, and Atlassian account IDs.
  • Not stored: raw secret or personal data values, prompts and AI responses, display names and email addresses of users, IP addresses.
  • Retention: closed and hidden findings 395 days by default (90 to 731 days on Standard, up to 2,557 days on Advanced), alerts 90 days, audit log 395 days, evidence packs 7 years.
  • Closed Atlassian accounts: the account ID is replaced by a pseudonym in every LeakRadar record. Evidence files that were already signed are not rewritten.
  • No data leaves Atlassian. See Privacy, Security and the DPA.

Editions

LeakRadar is licensed through Atlassian Marketplace with one licence for Jira and Confluence. The edition comes from the licence and is checked on the server for every action.

  • Standard: scanning, detectors, AI triage, triage, remediation, policies, DSAR and erasure, the weekly in-app digest, up to 20 custom detectors, OCR for 2,000 images per UTC month and retention from 90 days to 24 months. See Licence and Standard limits.
  • Advanced: everything in Standard. Included in Advanced: the Rovo agent, signed evidence packs, recurring weekly or monthly compliance packs, real-time alerts for new critical findings, up to 100 custom detectors, OCR with no monthly cap and retention up to 7 years. The home page pricing matrix shows both columns.

Known limitations

  • LeakRadar cannot check whether a secret is still active, and apps cannot purge Jira field history.
  • Roles and the LeakRadar team are granted to people, not groups. The team holds up to 50 people.
  • The work item panel and the byline are switched on or off together.
  • No precision or recall figures are published yet; see Detection methodology.
  • Evidence files signed before an account was closed are not rewritten: a findings CSV keeps that owner's account ID until the pack is deleted after 7 years.
  • A narrower work item type list applies to new scans only, because findings do not store the type.
  • Contextual detection of names and addresses is not included. Group member counts and guest lists are not read.
  • Comment backfill reads up to 20 pages of 50 comments per item and continues on the next scan.
  • The evidence findings CSV holds up to 200 rows. A recurring pack missed on its day is not created later.
  • The Open leaks gadget reads up to 50 projects.

FAQ

Do we need two licences for Jira and Confluence?

No. LeakRadar requires Jira and works with Confluence. Connecting Confluence uses the same licence.

Does LeakRadar store the secrets it finds?

No. It stores a masked snippet, a keyed hash and the location. The full value is never written to storage.

Does our content go to an external AI service?

No. AI triage uses Forge LLMs, which run inside Atlassian. For an ambiguous candidate the model sees up to 120 characters of text on each side, at most 2,000 characters per request, with every detected value replaced by a type, length and character set placeholder. AI triage is on by default and a Jira administrator can switch it off.

Can we limit AI usage?

Yes. On Built-in detectors set the monthly candidate budget (500 by default) or switch AI triage off. Over the budget, ambiguous candidates stay unclassified.

Can LeakRadar tell us whether a key still works?

No, because that needs a call to the provider outside Atlassian. Rotate every confirmed secret.

Can LeakRadar remove values from Jira history?

Apps cannot change Jira field history. LeakRadar flags history findings; rotate the secret and restrict the work item.

Will LeakRadar edit our content automatically?

Only if an administrator switches a policy live after its 7-day dry run. Otherwise every change is a confirmed action by a person.

What happens when our licence ends?

LeakRadar becomes read-only at once. Scans continue for 7 days and then pause. Findings and evidence stay available.

Can we send alerts to Slack?

LeakRadar creates Jira work items, shows alerts in the app and uses Jira notifications. Your own Jira Automation rules can forward those work items to Slack or Teams.

Troubleshooting

The scan paused by itself
A new LeakRadar version was installed. The next hourly reconcile resumes the scan, or resume it on the Overview.
The first scan is slow
The backfill runs newest first and in batches to respect Atlassian limits. Progress per product is on the Overview. Live events are processed as they arrive.
An item has errors
Coverage shows error counts per product as codes, never content. The hourly reconcile and the next full re-scan retry it; ask Support if errors stay.
A finding is a false positive
Mark it as false positive with a reason, and suppress the value so it does not come back.
Findings are counted but not shown
You cannot see the source item. Ask someone with access.
Findings have no AI label
AI triage is off, this month's candidate budget is used up, or the finding is definite and does not need one.
Buttons are disabled
The licence is inactive, so LeakRadar is read-only; the tooltip on the button says so. Check the licence under Manage apps.
Redaction failed
Redaction runs as you. Check that you can edit the item and that the item did not change since the scan.

Still stuck? See Support.

More Radrly apps for Jira