Daryll Gomas
Working tool · sample is fictional

Security audit

Every finding comes with its fix.

A read-only sweep of a small network, then a report an owner can read: a verdict, every finding ranked by severity, and a numbered fix for each. The next run can be compared with the last.

Sample · fictional business
Cover of the sample security assessment for a fictional dental office: verdict ACTION REQUIRED, 2 critical, 4 high, 6 medium, 1 informational
4stages in every sweep
2→1sweeps, Linux and Windows, one report format
0logins attempted: read-only
4severity levels, each with a deadline
13findings in the fictional sample
01The problem

A scan only helps if the owner can act on it.

A small office needs to know what is on its network and where the worst holes are: open ports, weak sign-ins, services with no password, what the internet can already see.

The person who approves the fix is usually not technical. A raw scanner dump doesn't reach them.

"What's on the network, what's here, and what are the worst holes … an executive summary, because you're going to be bringing this to people who are not technical."

02What I built

Two read-only sweeps, one report.

  • Linux sweep (Python, standard library only). Checks the perimeter, the machine it runs on, the local network and the public identity. Never logs in, sends no passwords, needs no root, changes nothing.
  • Windows sweep (the older PowerShell version). The same network checks, plus Windows endpoint checks: firewall profiles, disk encryption, antivirus, accounts, remote desktop, SMB.
  • Report builder (Python). Reads a run file from either sweep and writes one branded report: verdict, findings by severity, a numbered fix for each, and an optional "since last time" section.
  • A public sample for a fictional dental office, so the format can be shown without exposing a real network.
Which tool does what

The sample's endpoint findings (SMB1, BitLocker, antivirus, remote desktop, accounts) are Windows checks. The Linux sweep does not run them; the Windows sweep does.

Both sweeps write the same run-file shape, so one report builder reads either. Porting the Windows endpoint checks to the Linux sweep is not done yet.

03How it works

Read-only checks in, numbered fixes out.

How the security audit works: a read-only sweep (Linux or Windows) checks the perimeter, the machine it runs on, the local network and the public identity; each finding is scored and saved to one run file; the report builder turns it into a verdict and numbered fixes; after fixing, the next run is compared with the last. 1 · COLLECT, READ-ONLY LINUX OR WINDOWS SWEEP 2 · SCORE 3 · SAVE 4 · REPORT 5 · ACT NEXT RUN · COMPARED WITH THIS ONE Perimeter PUBLIC ADDRESS · EXPOSURE This machine OPEN PORTS · FIREWALL · DISK Local network PORTS · SSH · OPEN SERVICES Public identity DOMAIN LOCKS · PUBLIC DNS Score SEVERITY + ID Run file ONE JSON PER RUN Report VERDICT FROM WORST FINDING FINDING → NUMBERED FIX ONE SELF-CONTAINED FILE plus: since last time Fix NUMBERED PROCEDURES Re-scan COMPARE WITH LAST RUN How the security audit works: a read-only sweep (Linux or Windows) checks the perimeter, the machine it runs on, the local network and the public identity; each finding is scored and saved to one run file; the report builder turns it into a verdict and numbered fixes; after fixing, the next run is compared with the last. 1 · COLLECT, READ-ONLY LINUX OR WINDOWS SWEEP 2 · SCORE 3 · SAVE 4 · REPORT 5 · ACT Perimeter PUBLIC ADDRESS · EXPOSURE This machine OPEN PORTS · FIREWALL · DISK Local network PORTS · SSH · OPEN SERVICES Public identity DOMAIN LOCKS · PUBLIC DNS Score SEVERITY + ID Run file ONE JSON PER RUN Report VERDICT FROM WORST FINDING FINDING → NUMBERED FIX ONE SELF-CONTAINED FILE plus: since last time Fix NUMBERED PROCEDURES Re-scan COMPARE WITH LAST RUN
Animated diagram. Each box is a step in the code. The sample report is for a fictional business; no real scan is ever published.

1 · Collect: what each sweep checks

Stage
Linux sweep
Windows sweep (older)
Perimeterwhat a stranger sees
Linux sweepPublic address, a public exposure-database lookup, a local probe of open ports (marked indicative, not proof).
Windows sweepSame network check.
This machinewhere the sweep runs
Linux sweepListening ports, firewall service state, disk encryption. No root, so anything it can't read is marked unverified.
Windows sweepFirewall profiles, BitLocker, antivirus, local accounts and admins, remote desktop, SMB1 and signing, listening ports.
Local networkevery known host
Linux sweepPort scan against a stored baseline (new ports flagged NEW). SSH sign-in methods listed without logging in. Services checked for answering without a password (read-only requests).
Windows sweepSame network checks.
Public identitydomains and DNS
Linux sweepDomain transfer lock and registration privacy. Public DNS checked for internal addresses leaking out.
Windows sweepSame check.

The Linux sweep is a port of the Windows sweep's network checks. The Windows endpoint checks stay in the PowerShell version for now.

2 · Score

  • Every finding gets a severity and an ID
  • Sorted worst first
  • Critical: fix today. High: this week. Medium: this month.

3 · Save

  • One JSON run file: findings, counts, evidence
  • Never overwritten: each run keeps its own file
  • What couldn't be checked is listed, not hidden

4 · Report

  • Each finding ID maps to a numbered fix and its action
  • One self-contained HTML file: no outside requests
  • Refuses to overwrite an earlier report

5 · Fix & re-scan

  • Fixes follow numbered procedures, each with its own check
  • Next run compared with the last: fixed, still open, or can't check
  • Plus a then-to-now scoreboard and what's new

The verdict is a rule, not a judgment call

ACTION REQUIREDany critical finding
NEEDS WORKno critical, any high
FAIRonly medium or lower
GOODno findings against the baseline
04Tools

Small, plain, no installs.

Python 3standard library only: Linux sweep + report builder
PowerShellthe older Windows sweep
Linux built-insss, lsblk, systemctl: no root needed
OpenSSH clientlists sign-in methods, never logs in
Public exposure lookupwhat the internet already sees
RDAP + public DNSdomain locks, privacy, DNS leaks
Self-contained HTMLsystem fonts, no outside requests, prints cleanly
Claude Code, CodexAI writes the code; I specify and verify
05Proof

Pieces of the sample report.

Straight crops of the public sample. The business, addresses and machine names are made up.

Sample report, Critical findings table: F1, router admin page reachable from the internet on 8443, with the action to disable remote management; a remote desktop finding with the action to require Network Level Authentication
Swipe to read →Report builderA finding and its fix, side by side. The fix column holds the procedure number; "—" means no numbered procedure yet.
Sample report, Internal network: open ports per host with NEW ports in red, and an SSH table showing a host that still accepts password sign-in
Swipe to read →Network checks · both sweepsDrift against the baseline. Red ports weren't there last time. SSH shows a host that still accepts passwords.
Sample report, Endpoint: Windows 10 out of support, disk encryption off, antivirus inactive, an account with no password, remote desktop enabled, SMB1 enabled; host firewall profiles with logging off Sample report, Method: read-only collection, Windows posture via native management APIs
Swipe to read →Windows sweep onlyThe endpoint section is Windows data. The sample's own Method line says so: "Windows posture via native management APIs".
Sample report, What this assessment does not cover: account takeover, credential reuse and SIM swap, with seven items to confirm by hand each cycle
Swipe to read →Report builderIt says what it can't see. Seven checks that need a person, every cycle.

The sample is a single run, so it has no "since last time" section; that comparison is built into the report builder (step 5 above). Real reports hold real network details and stay private. Open the full sample ↗

06How it's tested and run

Safe to run, honest about its limits.

  1. Permission firstField work needs written permission from the owner of the network and devices before any scan.
  2. Read-only by designNo logins, no passwords sent, no root, nothing changed. SSH is asked which sign-in methods it allows, and that's all.
  3. Nothing overwrittenEach run writes its own file, and the report builder refuses to replace an earlier report.
  4. Unknown is not "fine"Checks it couldn't make are listed as unverified. In a comparison, items the new run can't see are marked can't-check, never fixed; a service that stopped answering is flagged to confirm, not counted as secured.
  5. Expected alarmsThe sweep trips intrusion detection. The report says so: an alert that matches a run is expected; one that doesn't should be investigated.
  6. Used on my own network firstThe Linux sweep was built to re-run my own audit and compare it with an earlier run. Those reports stay private.
07What I'd do next Proposed · not built

One sweep, a plain-words front page.

  • Port the Windows endpoint checks, so one sweep covers both.
  • Add an executive summary page for owners: the verdict, the top risks, what to do first.
08Skill tags
Security Security assessmentsPythonPowerShellLinux WindowsDNSClaude CodeCodex