Code and architecture audits that show their working.
We read your repository, never run it, and report what is wrong with the evidence beside each finding. Then we help you design what comes next, move off legacy in measured stages, and keep watch.
Trial: Audit Services, Active Support on one project, free. Engagements are scoped by the founder, Paramveer Singh.
Finding · SAST Specimen
Shell command built from a variable
src/legacy/export.ts
- Severity
-
Highgraded to Low - Why
- No entrypoint reaches this module.
- Kept
- The original severity, beside the new one.
- Status
- In the list. Not suppressed.
- Tool
semgrep 1.177.0- Rule
detect-child-process- Graded by
- ArchitectZero, against the repository
Four services, one evidence base.
Audit what you have, design what comes next, modernize it in measured stages, and keep it under watch. Pricing lists what each plan includes.
-
What is wrong with the codebase, with the evidence for each finding.
- You get
- A report as PDF or Word that opens with what the review covered and what it did not.
- Covers
-
-
Code review graded against the repository
Review lanes over the whole tree. In JavaScript, TypeScript, Python, Go, Java, Kotlin, C#, PHP, Ruby, Rust and Elixir, a finding is graded by whether a detected entrypoint reaches its file and how many files import it. Package advisories are graded by use for npm only.
-
A C4 architecture model and a 14-dimension audit
A C4 model derived from the tree, dependency cycles and drift, custom rules, and an audit scored on 14 due-diligence dimensions.
-
Secrets, SAST, IaC, workflow and container hardening
Secrets (gitleaks), SAST (Semgrep), IaC (Trivy), GitHub workflows (zizmor), container and Kubernetes hardening, and a threat-led rule pack for the paths attackers use.
-
Dependencies across 8 ecosystems, with EPSS
Advisories transitive through lockfiles, with EPSS, an SBOM and licences. A package we could not assess is counted as unassessed, never as clean.
-
Cloud and web reviews
Read-only review of AWS, Azure, GCP, Cloudflare and Railway estates for posture and cost. Accessibility, SEO and lab performance of a site you control.
-
SOC 2 and ISO 27001 readiness
126 SOC 2 and ISO/IEC 27001:2022 controls, assessed with a gap plan. Readiness, not certification.
-
- Not yet
-
Pentesting, DAST and API testing. Codebase grading beyond the eleven languages we index. Dependency-usage grading beyond npm.
-
What to build next, and why, written down so it can be reviewed and approved.
- You get
- Documents exported to Word or PDF from the text that was approved. Anything a model drafted is labelled; approval is a person’s act.
- Covers
-
-
Solution documents, versioned
Solution architecture, high- and low-level design, ADRs, API, data, and deployment and operations designs. Every unwritten section is shown as unwritten.
-
Requirements with MoSCoW coverage
Requirements and drivers with MoSCoW priority, and coverage that tells “uncovered” apart from “uncovered and a must”.
-
Decisions scored against weighted criteria
No ranking is shown until every cell is scored, and a decision that goes against the score names who decided and why.
-
A target architecture and the work it implies
A disposition for every discovered component, and the work it implies, in waves ordered by the target’s own dependencies.
-
- Not yet
-
Infrastructure or load simulation. Cost figures for a rewrite.
-
Off the legacy estate in stages, each one measured rather than declared done.
- You get
- A tracked programme. Before and after are two stored snapshots, so no stage is marked done by editing a number.
- Covers
-
-
The landscape, discovered from the code
Components and their dependencies, found in the tree’s own manifests and compose files.
-
Extraction order ranked by coupling
Stages with refactoring risk, effort and cost, at your hourly rate when you give one.
-
Each stage measured before and after
Progress is reported against the plan and against the whole estate.
-
Audit, target and schedule in one view
It names work the target calls for that no stage has claimed.
-
- Not yet
-
Automated code transformation. Grouping into bounded contexts. Bottlenecks from runtime data.
-
The same project, kept under review once the audit is done.
- You get
- Findings where your team already works: on the pull request, in your channel, on your status page.
- Covers
-
-
Pull-request review on GitHub
Check runs, a summary and inline comments, from the same lanes as the audit. Dispute a finding and it leaves the merge gate without being deleted.
-
Alerts to Slack, Teams or your pager
Rules by event, severity and category, sent to Slack, Microsoft Teams, a webhook, PagerDuty, Opsgenie or ntfy, for the workspace or one project.
-
Site health and TLS expiry
HTTP, browser-rendered and heartbeat monitors of a site you have proved you control, with maintenance windows. Latency is synthetic and labelled synthetic.
-
A public status page
For the monitors you choose to publish, with posts you write.
-
- Not yet
-
Fixes written to your branch. Runtime metrics from your own systems.
How an audit runs.
From a repository to a report, without anything in it being run.
-
Connect a repository.
Install the GitHub app on the repositories you choose, or upload an archive. An archive is virus-scanned before anything reads it.
-
Read, never run.
Secrets, SAST, IaC, workflow and dependency scanners run in a container with no network. Findings in the languages we index, and npm advisories, are then graded against the repository.
-
Read the report.
PDF or Word, opening with what the review covered and what it did not. Every finding names its tool, version and rule.
Graded with a reason, never hidden.
- Graded down, with the reason.
- In JavaScript, TypeScript, Python, Go, Java, Kotlin, C#, PHP, Ruby, Rust and Elixir, a finding in a file no detected entrypoint reaches is graded down, and the report says why. Reachability is read per file, and an entrypoint we miss can grade a live file down.
- The original severity stays.
- It is printed beside the new one, so anyone can check the change.
- A dispute leaves the gate, not the record.
- Disagree with a finding on a pull request and it stops blocking the merge. It is not deleted.
- Unassessed is not clean.
- A dependency we could not assess is counted as unassessed.
How your code is treated.
Stated plainly, so you can hold us to it. The controls behind each line are on the security page.
- Isolated per workspace.
- Postgres row-level security, forced for the application’s database roles.
- Analysed as data, never run.
- gitleaks, Semgrep, Trivy, zizmor and Syft run in a container with no network, under a user of their own.
- Model use is yours to switch off.
- On by default, per workspace or project, and every run records which way it was set. Excerpts are masked before they leave.
- Never used for training.
- No model is trained on your code, your findings or your reports.
- Kept only as long as it is needed.
- Clones are deleted on a schedule, archives when you ask. Findings are kept for the ten most recent runs.
- Every finding shows its source.
- A scanner finding carries the tool, its version and the rule; one of ours names the analyzer that raised it.
SOC 2 Preparing for Type I against the Security criteria. No audit has been completed yet. Where we stand
What it does not do.
Better to read it here than to find it in the report.
Not offered
- Pentesting or dynamic application testing.
- On-call or incident response by us. Alerts reach your pager, and your team answers it.
- Fixes applied to your branch.
- Certification. Readiness against 126 controls is not SOC 2 or ISO compliance.
Not built yet
- Codebase grading beyond the eleven languages we index.
- Dependency-usage grading beyond npm.
- Load simulation, or modernization that rewrites code for you.