Find your crown jewels in Microsoft 365, Google Workspace or AWS, see how exposed they are, and hand leadership a branded PDF that explains the risk.
crownguard is a guided self-assessment that runs entirely in your browser. You pick your environment, name the systems and information that would hurt most if compromised (your Entra ID tenant, Global Administrators, the HR SharePoint site, the finance shared drive, Gemini or Copilot's reach into your data), answer plain-language control questions drawn from Microsoft, Google and AWS security guidance, and download a report themed with your organisation's logo and colours.
Nothing you enter leaves your device. There is no backend, no account and no analytics.
Try it: https://jusso-dev.github.io/crownguard/
- Crown-jewel discovery. Guided prompts per category: identity plane, privileged access, business data, collaboration and email, endpoints, cloud infrastructure and AI assistants.
- Vendor-grounded questions. Each question cites the Microsoft, Google or AWS guidance it comes from and maps to CIS Benchmark recommendations, the ASD Essential Eight, ASD Essentials (enterprise IT and cloud), NIST CSF 2.0, the Australian Government Information Security Manual (ISM, September 2026 release) and the Department of Home Affairs Industry Data Classification Framework (IDCF).
- Cloud posture for Azure, Google Cloud and AWS. Optional Azure and Google Cloud modules and an AWS platform ask about the misconfigurations a cloud security posture management (CSPM) tool flags first: privileged access to the management plane, public storage, databases and snapshots, secrets in configuration, logging and threat detection, organisation guardrails and backups an attacker can't delete. They map to the CIS Foundations Benchmarks for each cloud, the Microsoft cloud security benchmark and AWS Foundational Security Best Practices.
- Risk per crown jewel. Impact × likelihood on a 5×5 matrix, driven by your answers, the exposures you record and how sensitive or regulated the data is.
- A report people will read. It opens by stating the scope, the standards it's assessed against and that it is a point-in-time assessment, then an executive summary, heatmap, crown-jewel register, risk register, findings with fixes and licence notes, controls marked N/A with the reason given, a 30/60/90-day roadmap, indicative Essential Eight maturity, IDCF Data Security Levels, NIST CSF coverage and references. Your logo, ABN, colours and protective marking, with the generation time and the time answers were last changed stamped on every report.
- Save and come back. Progress saves automatically in your browser as you go. Return later and you pick up
on the same step and section. Save file (or Ctrl/⌘ S) writes a
.crownguard.jsoncopy without moving you off the question you're on; in Chrome and Edge later saves update the same file. Open file carries on from a saved copy, including older ones: open it, add a logo or ABN, and regenerate the report. - Optional scan import. Already run a scanner? Import its results on the Controls step to pre-fill answers where its
checks are decisive (all pass = Yes, all fail = No, mixed = Partial; a pass alongside a check the scan couldn't
settle, such as one needing review, is Partial). Today that covers M365-Secure
for Microsoft 365 (
_Assessment-Results_<domain>.json, 59 questions) and Prowler for AWS (56 questions) and Azure (33 questions), as CSV or JSON-OCSF straight fromprowler aws -M csv json-ocsf. Prowler reports one finding per resource, so a check fails when any resource fails it and a suppressed finding is never a pass; the evidence carries the counts ("3 of 41 resources failing") and example resources. ScubaGoggles for Google Workspace (ScubaResults*.json, 33 questions) is read the same way from its policy results only: the report'sRawsection, which holds super admin and break-glass accounts and audit log events, is never touched. Account email addresses in any scanner's text are replaced with "[account]" before anything is stored. Every pre-filled answer shows its scan evidence, you can change any of them, and the report names the scanner, its version and the account it covered. An answer you keep keeps its own evidence: a later scan's checks never sit under it. The file is read in your browser only. - N/A must be justified. Marking a control N/A asks why. Until a reason is given it counts as unanswered.
- IDCF Data Security Levels. Optionally record the IDCF Data Security Level (DSL-0 to DSL-5+) your data owner chose for each crown jewel. The report shows, for each platform and for DSL-2 to DSL-4, whether your answers found gaps in the physical, cyber and authorised-person protections the IDCF describes (the cyber part comes from indicative Essential Eight maturity, as the IDCF does), and checks each tagged crown jewel against its level. It never assigns levels for you or claims a level is met.
- Optional SOC maturity assessment. If you have a security operations centre, in house or through a managed provider, rate it on 37 plain-language questions structured on the five domains and 27 aspects of the SOC-CMM® v2.4 model. Each question describes what levels 0 to 5 look like (0 to 3 for technology and service capability), so a rating means the same thing every time. The report adds a radar against targets, an aspect profile, and what the next level up looks like for the eight largest maturity gaps and every capability gap. It is indicative, reported separately and never changes crown-jewel risk.
- Optional AI use-case register. Record each AI use case or agent (Microsoft 365 Copilot, Copilot Studio agents, Gemini, ChatGPT and other public tools, third-party AI connectors and OAuth apps, and your own agents) with the minimum register fields in the Digital Transformation Agency's Standard for accountability, mostly by clicking options. Each use case gets up to 22 readiness questions on accountability, human oversight, access, data, monitoring, switch-off, transparency and assurance, drawn from the Policy for the responsible use of AI in government v2.1, its impact assessment guidance, the agentic AI addendum and ASD's guidance on agentic AI. Link a use case to the crown jewels it can reach to see their risk, the related Controls answers and the exposures it usually adds. Export the register as CSV or XLSX, with the DTA's fields first, and the report gets a register section. Example entries are available, and they're labelled as examples everywhere. The register also runs on its own as a five-step AI register only flow with its own PDF.
The screenshots below follow one assessment for a fictional health provider, Riverbend Health, on Microsoft 365. They
are regenerated from the app itself (SCREENSHOTS=1 pnpm exec playwright test e2e/screenshots.spec.ts).
Start with the organisation: name, ABN (checked against the ATO check digit), sector, size and the obligations that apply to it, such as the Privacy Act and its Notifiable Data Breaches scheme. A logo is optional and sets the report's colours. Then choose the platforms in scope (Microsoft 365 and Entra ID, Google Workspace, Amazon Web Services, or any mix), with their licence tiers and optional Azure or Google Cloud modules.
![]() |
![]() |
Guided prompts per category (identity plane, privileged access, business data, collaboration and email, endpoints, cloud and AI assistants) help you name what would hurt most. Each crown jewel gets a classification, confidentiality, integrity and availability ratings, the exposures that apply today (standing admin rights, guest access, unmanaged devices, AI grounding), its obligations, and optionally the IDCF Data Security Level your data owner chose. crownguard can suggest a level, but never fills it in for you.
Only the questions relevant to your crown jewels and licences are asked, grouped by domain. Each one explains why it matters, what "yes" looks like and how to fix it, names any licence the fix needs, and lists the Microsoft, Google or AWS pages it comes from plus its CIS, Essential Eight, NIST CSF 2.0 and IDCF mappings. Unknown counts as a gap; N/A needs a reason. Answers can be pre-filled from a scan you have already run: M365-Secure for Microsoft 365, Prowler for AWS and Azure, or ScubaGoggles for Google Workspace.
If you have a security operations centre, in house or through a managed provider, rate it across the five domains and 27 aspects of the SOC-CMM® v2.4 model. Every question describes what each level looks like, so you pick the description that matches today rather than a vague "partially". Technology and service aspects also get a capability rating, aspects you genuinely don't need can be left out of scoring, and the targets default to SOC-CMM's (maturity 3, capability 2).
Add each AI use case from a preset, which fills in the usual technology type, usage pattern and access. Then record the register fields the DTA's Standard for accountability lists: name, reference, description, technology type, lifecycle stage, use of the AI technical standard, domain, usage pattern, accountable owner, the Appendix C criteria that put it in scope, the risk ratings from its AI impact assessment and, for high-risk use cases, its review dates. crownguard records the risk ratings but never works them out. The oversight model (output only, human in the loop, human on the loop, human out of the loop), access level and data handled decide which readiness questions apply, so agent questions are only asked of AI that can take actions. The screens below use the example entries.
Link a use case to the crown jewels it can reach and you'll see each one's current risk, your Controls answers on the same tool, and the exposures it usually adds (AI grounding, third-party app access). Ticking a suggested exposure updates the crown jewel and its risk; nothing changes unless you tick it. The readiness panel shows each theme's status and every open gap, most severe first.
![]() |
![]() |
Download CSV and Download XLSX export the register with the Standard's 16 fields first, in its order (the owner's name and email are split into two columns), then crownguard's columns: example flag, type, product, oversight, access, data, linked crown jewels and readiness. The XLSX adds an About sheet with the key dates, caveats and sources. Both are built in the browser.
Risk for each crown jewel is impact × likelihood on a 5×5 matrix, recalculated as you answer. The Review step shows control posture, answer confidence, the heatmap, crown jewels ranked by risk, posture by domain, indicative Essential Eight maturity and the biggest gaps, plus the SOC maturity result and the AI register when you included them. The optional "Show ASD Essentials" picker adds indicative Essentials ratings for the enterprise IT or cloud chapter to the report, beside the Essential Eight.
![]() |
![]() |
Set who the report is prepared for and by, the protective marking, and the colours (taken from the logo, or typed as hex codes, with a contrast check). Then generate the PDF in the browser.
![]() |
![]() |
A report leadership will actually read: it opens with scope, the standards it was assessed against and a point-in-time statement, then an executive summary, the crown-jewel register, a risk register, findings with fixes, a 30/60/90-day roadmap, framework alignment (Essential Eight, ISM, IDCF, NIST CSF 2.0, CIS) and, when included, the SOC maturity and AI use-case register sections. Every page carries your marking and the time it was generated.
It works on a phone too: every step fits a 390px screen.
If you only need the AI use-case register — the one the Policy for the responsible use of AI in government asks non-corporate Commonwealth entities to create and share with the DTA every six months — you don't need to invent a platform or a crown jewel to get it. Choose AI use-case register only on the start screen, or open the deep link:
https://jusso-dev.github.io/crownguard/#/ai-register
The flow is five steps — organisation, AI register, review, branding, report — and the only thing it needs before the register is an organisation name. Crown-jewel links are left out; switching to the full assessment later adds them without losing anything.
You get the same register exports as always — CSV and XLSX with the Standard's minimum fields first — plus a standalone AI use-case register PDF: a branded cover, an "About this register" page with who the policy applies to and a point-in-time statement, the register with readiness for each use case, the key dates, where the sources are unclear, and the sources themselves. It carries no risk register, roadmap or framework pages: it's a document your agency can file or share.
When you're ready for the full assessment, the Report step has Turn this into a full crown-jewel assessment. It keeps every entry, answer and note, and takes you to the Environment step to name your platforms and crown jewels.
Hosted: https://jusso-dev.github.io/crownguard/ (GitHub Pages, built from main)
Or run it yourself:
pnpm install
pnpm dev # http://localhost:5173
pnpm build # static site in dist/, host anywhereThe build is a static site with no server code. Serve dist/ from any static host; set BASE_PATH when it lives
under a sub-path (the Pages workflow uses /crownguard/). The Content Security Policy ships as a <meta> tag
because GitHub Pages can't send custom headers; on a host that can, also send the headers defined in
vite.config.ts (including frame-ancestors 'none' and Cross-Origin-Opener-Policy, which only work as headers).
Copy-paste snippets: docs/self-hosting.md.
Nothing you enter is sent anywhere. Progress is kept in this browser under crownguard:v1 unless you choose otherwise:
- Optional passphrase — encrypts the autosaved copy and any file you save while it is set (WebCrypto PBKDF2-SHA-256 + AES-GCM). The passphrase stays in memory only for this visit; there is no recovery if you forget it.
- Don't keep this assessment in this browser — assessment data stays in memory only (not
sessionStorage). Reloading loses it. A small mode flag may remain so the choice survives a reload. - Remove from this browser — clears the storage key and drops the in-memory assessment immediately.
Browser storage is per origin. On GitHub Pages that origin is shared by every Pages site under the same account.
A dedicated origin is recommended but not configured yet (no custom domain has been chosen); production stays on
https://jusso-dev.github.io/crownguard/ for now. For sensitive assessments prefer a passphrase, no-persistence mode,
Save file, a private window, or self-hosting on your own domain. See docs/threat-model.md.
tui/ builds crownguard as a single terminal app (Ratatui) with the question catalogue
inside it: no browser, Node or network needed. It runs the crown-jewel assessment from organisation through to the
review (posture, heatmap, risks, Essential Eight and roadmap) and saves the same .crownguard.json file, so you can
switch between it and the web app at any point.
cargo build --release --manifest-path tui/Cargo.toml
./tui/target/release/crownguard riverbend.crownguard.jsonThe PDF, scan import, SOC maturity and the AI register stay in the web app. See docs/tui.md.
For consulting workflows, CI, or agents — validate, import scans, score, render PDFs and export the AI register from Node with no browser and no network:
pnpm cg new --org "Riverbend Health" --platforms microsoft --out a.crownguard.json
pnpm cg score a.crownguard.json
pnpm cg render a.crownguard.json report.pdfSee docs/cli.md for every command, exit codes, and how to replace the old Hermes cg-bot helper.
crownguard also comes as an offline desktop app for macOS, Windows and Linux, built with Tauri v2 from the same code. It keeps assessments in local files and a local SQLite database instead of browser storage, saves through the system's own dialogs (Ctrl/⌘ S saves back to the same file), and makes no network connections at all unless you turn on the optional update check. Files move freely between the web and desktop apps.
pnpm tauri:dev # run it locally (needs Rust and the Tauri prerequisites)
pnpm tauri:build # build installers for this systemSigned installers are published as desktop-v* releases. Install steps, where data lives, the offline guarantees and
how to back up assessments: docs/desktop/README.md. Threat model:
docs/desktop/security.md.
Save file writes one JSON object with the whole assessment. Nothing is sent anywhere: the file is your data, and you choose where it lives. The format is versioned, documented and published as a JSON Schema, so other tools can write and validate it — see docs/file-format.md.
A file saved by an older crownguard opens unchanged and is re-saved in the current format. A file saved by a newer
crownguard is offered read-only rather than quietly rewritten. Fields this version doesn't understand are carried
through a re-save untouched, and answers to questions that have since been retired are kept under orphans instead of
being dropped.
| Answers | Yes = 1, Partial = 0.5, No = 0. Unknown and unanswered also score 0: uncertainty is never treated as protection. N/A is excluded. |
| Weights | Critical 4, High 3, Medium 2, Low 1 |
| Impact (1–5) | Highest of the confidentiality, integrity and availability ratings, +1 for regulated or highly confidential data, max 5 |
| Likelihood (1–5) | 1 + 4 × weighted gap ratio of the questions that apply to the crown jewel on its own platform (its asset type plus platform-wide controls; another selected platform never counts), +0.5 per recorded exposure, rounded. At least 3 while any critical control is missing |
| Bands | 1–4 Low · 5–9 Medium · 10–19 High · 20–25 Extreme |
| Essential Eight | A maturity level counts only when every question at that level and below is Yes or N/A with a reason. Indicative only: cloud-platform controls, not a full ASD assessment |
| Roadmap | Gaps ranked by risk reduced across the crown jewels on that control's platform ÷ effort. Critical gaps with small or medium effort go into the first 30 days |
| IDCF | Cyber part of DSL-2, 3, 4 read from indicative Essential Eight Maturity Level 1, 2, 3; physical and authorised-person parts from the questions mapped to each level. A level shows gaps when any mapped question at or below it isn't Yes or N/A with a reason |
| ASD Essentials | Optional. Each action is rated per domain from the questions tagged with it: any No is not effective, fully effective needs every question Yes or N/A and one Yes covering the whole action, Unknown alone is not assessed. Principles take their weakest action; a domain's percentage is NE 0, PE 0.5, FE 1 over the principles assessed. Domains are ordered by ASD's priority phases, which set the order of work, not a rating. Never calculated from Essential Eight levels. Indicative, not a formal Essentials assessment; cloud chapter ids are provisional |
| ISM | Mapping only, generated from ASD's ISM OSCAL catalog (September 2026 release). The optional "Show ISM baseline" picker annotates findings that touch the NON_CLASSIFIED, OFFICIAL: Sensitive or PROTECTED baselines and adds a count; it never changes a score |
| SOC maturity | Each question rated 0–5 (capability 0–3) against its level descriptions; Unknown scores 0. An aspect's maturity is the mean of its maturity ratings and, for technology and services, its capability the mean of its capability ratings (never combined); a domain is the unweighted mean of its in-scope aspects. Default targets maturity 3, capability 2. Kept apart from the risk scores |
| AI readiness | Scored like the controls (Yes 1, Partial 0.5, No, Unknown and unanswered 0, N/A with a reason left out, severity-weighted) over the questions that apply to each use case. Agent questions are asked unless it only gives output. Indicative, not a DTA rating, and kept apart from the risk scores |
The code is in src/engine with tests alongside.
All questions, crown-jewel categories, framework mappings and citations are YAML in content/, so you
can improve them without touching code. See docs/content-guide.md and
CONTRIBUTING.md.
The optional SOC maturity questions live in content/soc/, which is licensed CC BY-SA 4.0 because
it builds on the SOC-CMM® model's structure (see its NOTICE).
The AI register lives in content/ai-register/: model.yaml holds who it applies to, the key
dates, presets, themes and the points the sources leave unclear, and questions.yaml the readiness questions. Each
question records whether its source says "must" or "should" and cites the DTA or ASD pages it comes from
(content/sources/dta.yaml). The agentic AI addendum's criteria and the AI technical
standard's criteria it maps to are in content/frameworks/.
crownguard says so in the app, the exports and the report rather than guessing:
- The agentic AI addendum describes itself as best practice to apply with the AI technical standard, and the policy strongly recommends that standard rather than requiring it. Its "must" and "should" are shown as written.
- The addendum asks for minimum technical requirements for kill switches but doesn't define them.
- The register's AI technology types have no agentic type, so agents are recorded under the technology they use.
- The policy gives the register deadline as "within 12 months" of 15 December 2025. The 15 December 2026 date shown is that arithmetic and is marked as worked out.
- Appendix B allows general-purpose AI such as Copilot to be one use case or several.
- The Standard lists the owner as one "name and email address" field; crownguard keeps them as two.
- The policy applies to non-corporate Commonwealth entities. Anyone else can use the register as a benchmark, but the dates and requirements aren't theirs.
Sources include Microsoft Learn (Zero Trust deployment guidance, Entra ID security operations, privileged access, Conditional Access, Microsoft 365 Copilot oversharing guidance, Purview, Defender), Google Workspace Admin Help (security checklists, administrator and super admin best practices, Drive, Gmail, Gemini) and Google Cloud (enterprise foundations, organisation policies, Security Command Center), Azure documentation (Azure RBAC and PIM, Defender for Cloud, Storage, Key Vault, networking, Azure Backup), AWS documentation (IAM and IAM Identity Center, Organizations, S3, CloudTrail, GuardDuty, Security Hub CSPM, AWS Backup), plus the ASD Essential Eight Maturity Model, the Australian Government Information Security Manual (from ASD's ISM OSCAL catalog), NIST CSF 2.0, the Home Affairs Industry Data Classification Framework, the SOC-CMM® model, the Digital Transformation Agency's AI policy, standards, impact assessment guidance and agentic AI addendum, and ASD's guidance on agentic AI.
Vendor guidance moves, changes and retires. A nightly workflow (source-watch.yml)
re-reads every cited page and compares it with the fingerprints in watch/state.json. It opens a
pull request when a source:
- breaks: the page is gone, or fails on consecutive nights;
- moves: the same document now lives at a new address, so the PR updates the URL for you;
- redirects elsewhere or retires: it lands on a hub page, is archived, or is marked retired or "classic";
- changes substantially: sections added or removed, a large share of the text rewritten, or new deprecation or licensing language (with a diff excerpt where the page's licence allows quoting);
- has a newer version: for example a new CIS Benchmark release;
- passes a deadline: a date the page gives for a retirement or switch-off has arrived;
- changes with the rest of its site: when many pages on one site change on the same night (usually new page furniture), they are listed once as a likely template change rather than one finding per page;
and when new guidance appears: new pages next to a cited page in Microsoft Learn's tables of contents, security and
admin posts on Google Workspace Updates, deprecations or security notes in Google Cloud release notes, and ASD, CIS
and NIST announcements. Merging the PR accepts what it reports; anything still unfixed is listed under "Still
unresolved" in later PRs without reopening one on its own. Page text is never committed: fingerprints are hashes,
section names and word counts, and the text needed for diffs stays in the Actions cache. pnpm validate:content
separately rejects sources that nothing cites. See Source watch.
A second nightly workflow (guidance-update.yml) regenerates the ISM controls
and the Prowler and ScubaGoggles check id lists when upstream publishes a new release, and opens a pull request with
the changes. See Guidance update.
CIS Benchmarks are referenced by recommendation number and title only, under CIS's CC BY-NC-SA 4.0 terms. Get the benchmarks from CIS for full audit and remediation steps.
crownguard targets WCAG 2.2 Level AA. Every step is checked for it automatically: axe on every wizard step in Playwright, plus keyboard, focus, target size and reflow tests, and a unit test that keeps the palette's contrast above AA. Known limitation: the PDF report is not tagged for screen readers yet — use the on-screen Review step with assistive technology. See docs/accessibility.md for the manual checklist and results. Report a problem via SECURITY.md or a GitHub issue.
This is a self-assessment, not an audit. It doesn't connect to your tenant or verify answers, and it covers your Microsoft, Google or AWS cloud platform, not your whole environment (on-premises networks, line-of-business apps, suppliers, physical security). Use it to start the right conversation and to prioritise, then verify.
The AI register's readiness check is a self-check against the DTA's published policy and guidance, not a compliance finding or a DTA assessment, and it doesn't replace your AI impact assessment.
Vite, React, TypeScript, Tailwind CSS, Zustand, Zod, @react-pdf/renderer, Vitest, Playwright. The desktop app adds
Tauri v2 and SQLite (rusqlite) in src-tauri/. The terminal app is Rust with Ratatui in tui/. Fonts (all SIL
Open Font License, self-hosted): Space Grotesk, Inter and JetBrains Mono in the app via @fontsource; Inter in the
PDF (src/report/fonts/OFL.txt). The UI design system is documented in design.md.
The code is MIT (LICENSE). The SOC maturity content in content/soc/ is
CC BY-SA 4.0 because it adapts the structure of the SOC-CMM® model
by Rob van Os (NOTICE). The IDCF content paraphrases the Industry Data Classification
Framework, © Commonwealth of Australia 2026 and © Commonwealth Scientific and Industrial Research Organisation (CSIRO)
2026, licensed CC BY 4.0 (the Coat of Arms excepted), and "IDCF: A
guide to system security" by the Australian Government Department of Home Affairs, used under the department's website
terms (CC BY 3.0 AU). The ISM control titles in content/frameworks/ism.yaml are
shortened statements from the Information Security Manual, © Commonwealth of Australia 2024, generated from ASD's
ISM OSCAL catalog and licensed CC BY 4.0 (the Coat of Arms and ASD
logo excepted). The ASD Essentials action titles in content/frameworks/asd-essentials-*.yaml are shortened
statements from ASD's Essentials for enterprise IT (generated from ASD's E4EIT OSCAL release) and Essentials for cloud,
© Commonwealth of Australia 2026, licensed CC BY 4.0 (the Coat of Arms
and ASD logo excepted). The AI register paraphrases Digital Transformation Agency policy, standards and guidance,
© Commonwealth of Australia, licensed CC BY 4.0, and draws on ASD
guidance. CIS Benchmarks are referenced by recommendation number and title only, under CC BY-NC-SA 4.0.
crownguard is independent and not affiliated with or endorsed by Microsoft, Google, Amazon Web Services, CIS, ASD, NIST, the Department of Home Affairs, CSIRO, the Digital Transformation Agency or SOC-CMM. SOC-CMM® is a registered trademark of its owner. Product names are trademarks of their respective owners.



























