Skip to content
CyberArmorApplication security engineering
São Paulo, Brasil · The platform

One platform for the whole assessment, from raw scanner output to the signed report.

It reads five tool output formats, gives a finding an identity that a reformat does not break, suggests each finding to a control in three loaded catalogues, and drafts the executive report. Every count on this page was read in the engine source, and every one of them arrives with its boundary.

SARIFCYCLONEDXNESSUS XMLCSV

Five engines, and the edge of each one

198code rules across 14 languagesTaint tracking on JavaScript and TypeScript only, inside a single function, off by default. Elsewhere it is line-by-line pattern matching, and one line cannot see the one before it.
8dependency ecosystems, 9 manifests and 12 lock files, with the transitive treeNo operating-system package is read here. That only comes from inside a container image.
31secret patterns, plus 183 IaC rules and 100 container rulesIaC coverage is lopsided: 111 rules for AWS, 16 for Azure, none for Google Cloud.

Read item by item in the engine source. No count here was rounded.

What it reads

Your tools stay. The platform reads them.

Five readers, and everything they produce lands in the same findings table as what our own engines found.

  • SARIF 2.1.0

    CodeQL, Semgrep, SonarQube, Fortify, and anything else that emits SARIF. On the way back, it exports SARIF.

  • CycloneDX

    Trivy natively, and Grype.

  • Nessus XML

    Nessus and OpenVAS.

  • DAST output

    ZAP, Burp and Nuclei.

  • Vendor CSV

    Qualys, Rapid7 and Tenable.

Import ships behind a feature flag that is off by default. Until it is switched on in the environment, those endpoints answer 404. It is on the configuration list, and it is the kind of thing you would find out on day one.

What happens after a finding arrives

A finding you accepted stays accepted

This is the part that decides whether a retest costs you an afternoon or a week.

  • Identity ignores line numbers

    It is the rule, the relative path, the normalised snippet and an index. Reformatting a file does not invent a new finding.

  • The same flaw from two scanners is one finding

    A vendor-independent fingerprint joins them, so you get one finding with two sources instead of two findings to reconcile by hand.

  • Bulk marking refuses to close a critical

    Mass triage will not close a critical or high finding. The one action that would quietly hide the thing you are paid to find is blocked in code.

Compliance

Three catalogues loaded — which is not the same as three certifications

The distinction matters more than the number, and it is the first thing an auditor asks. These are catalogues inside the product. We hold no certification.

93ISO/IEC 27001:2022 controls, counted one by oneA catalogue in the product. Not a certificate we hold.
47SOC 2 criteriaA catalogue in the product. There is no auditor report to attach.
90OWASP SAMM v2 activitiesMaturity assessment built in.
  • It suggests. You classify.

    A finding is matched to a control by keyword, and the confidence of that guess is recorded on the record. You confirm or reject it. The platform does not certify anything.

  • Evidence expires

    Evidence carries a date. Expired evidence stops supporting its control instead of sitting in a spreadsheet as if it still counted.

  • Compliant and assessed are two different numbers

    The screen keeps them apart. Merging them is how a coverage figure lies, and this house has paid for that lesson.

The report

The last mile, which is what you actually sell

Finding the flaw was never the bottleneck. Writing the document was.

  1. The questionnaire adapts to the client profile

    A different profile gets a different set of questions, so you are not deleting irrelevant sections by hand.

  2. The score is computed in code

    Not assembled in a spreadsheet, and not different depending on who filled it in.

  3. The executive report comes out drafted

    Six fixed sections: summary, risks by impact, what is in good shape, and a 90-day plan. You review, correct and sign.

  4. It exports as PDF

    And only as PDF. Not Word, not Excel.

  5. One record per client company, with its own history

    Each client keeps its own assessments, scores and reports, under your account.

Reads / stops at

Where each engine ends

The same table we would have to walk you through in week two.
EngineReadsStops at
Code198 rules across 14 languagesTaint on JS and TS only, inside one function, off by default
Dependencies8 ecosystems, lock files, transitive treeNo operating-system packages
Containers100 rules, dpkg and apk packagesRPM and RedHat are out of scope, stated in the source
Infrastructure as code183 rules111 AWS, 16 Azure, no Google Cloud
Web scanningThe anonymous surfaceNo login, no certificate, cipher or TLS version inspection
Compliance93 ISO/IEC 27001:2022, 47 SOC 2, 90 OWASP SAMM v2Keyword suggestion with recorded confidence, not classification
Limits

What we do not do

A security vendor that claims to do everything is selling, not assessing. Every limit below was read in the code before it was written here, and it is in writing before the proposal, not after it.

  • Data-flow tracking is narrow

    From input to dangerous sink, it exists in JavaScript and TypeScript only, inside a single function, and it ships off. In other languages it is pattern matching, line by line, and one line cannot see the previous one.

  • The web scanner sees the anonymous surface

    It does not log in to your client system, and it does not inspect certificate, cipher or TLS version. When it confirms a flaw it stops there: it does not escalate privilege, run commands or extract data.

  • OS packages and cloud coverage are uneven

    RedHat and RPM packages generate no CVE. Debian, Ubuntu and Alpine do, and only from inside a container image. In infrastructure as code: 111 AWS rules, 16 Azure, none for Google Cloud.

  • Three repository providers, in the cloud

    GitHub, GitLab and Azure DevOps. Not Bitbucket, not self-hosted GitLab, not GitHub Enterprise Server. There is no pipeline Action and no command-line program: wiring it in is an HTTP call you write.

  • White label does not exist

    The report carries our brand in the header and footer of every page. There is no tenant field for a logo, a colour or a name — changing it takes a code change and a rebuild. It is first in the queue.

  • PDF only, and no portal for your client

    No Word, no Excel. And the company you assess does not log in to see its own risk — what it gets is the document you send.

  • The platform is not live yet

    Infrastructure is being rebuilt. There is no signup, no trial and no billing, and you pay for nothing before it exists.

Next step

Tell us which scanners you run and we will tell you what lands in the table.

A technical conversation with the person who wrote the code. No slide deck, and no proposal attached.

Talk to the engineer