Writing
All writing
Everything I have published, newest first.
RSS | 80 posts published | Original research: data studies
- Security Engineering 38
- AI Security 23
- MCP Security 23
- Engineering Practice 16
- Internet Governance 11
- Deployment 9
- DNS Security 8
- Security Operations 8
- Email Security 5
- Open Source 5
- Incident Response 4
- Security Automation 4
- Identity 3
- Data Privacy 2
- Integration Reliability 2
- Automation 1
- OSINT 1
-
7 min read
My MCP Scanner Scored 0 of 2 on a Real CVE. Here Is What Found It Instead
mcpscan 0.21.0 gave the vulnerable and the patched mcp-atlassian the same 7 findings and missed both CVEs. A manual trace found the file write. What 0.22.0 fixed and what it still misses.
- MCP Security
- Security Engineering
-
10 min read
Microsoft 365 Went Down Overnight. 17.6% of Indonesian co.id Domains Get Their Mail Through It
What Microsoft has said about the 12-13 November 2026 Exchange Online outage, in WIB, why restarts kept failing, and a DNS check of who receives mail for 3,499 .id domains.
- Security Engineering
- Email Security
- Internet Governance
-
6 min read
AI Agent Saya Bilang Kita Sedang Diserang. Log-nya Berkata Lain.
AI agent saya melaporkan prompt injection, tapi log tidak menunjukkan apa pun. Bagaimana pengecekan forensik membongkar alarm palsu, dan apa yang saya ubah.
- AI Security
- Security Operations
-
6 min read
Apa Itu Keamanan MCP? Serangan Umum dan Cara Memindai Server MCP Kamu
Apa itu keamanan MCP? Empat serangan dari pengujian 11 server MCP: tool poisoning, command injection, path traversal, planted prompt, dan cara memindainya.
- MCP Security
- AI Security
-
9 min read
A 1-Day Cooldown Would Have Blocked 70% of 2026's Hijacked npm Releases. Against Typosquats, 7 Days Still Misses 45%
I replayed 11,556 malicious npm and PyPI packages from 2026 against a dependency cooldown. Hijacks die fast, typosquats don't, and fixes pay the bill.
- Security Engineering
- Engineering Practice
-
4 min read
Cara Deploy Cloudflare Worker via GitHub Actions: Error 10000 Gara-gara API Token Salah
Deploy Cloudflare Worker pertama via GitHub Actions selesai 47 menit, 3 menit habis di error auth 10000 karena satu template API token. Setup Wrangler yang jalan.
- Deployment
- Engineering Practice
-
8 min read
Cooldown 1 Hari Memblokir 70% Rilis npm yang Dibajak di 2026. Untuk Typosquat, 7 Hari Masih Meloloskan 45%
Saya memutar ulang 11.556 paket npm dan PyPI berbahaya 2026 terhadap cooldown dependensi. Pembajakan cepat mati, typosquat tidak, dan perbaikan ikut tertunda.
- Security Engineering
- Engineering Practice
-
9 min read
DNS Publik Mullvad Tutup 2 November: 5 Pengecekan Sebelum Pindah ke Quad9
DNS publik Mullvad berhenti 2 November. Quad9 tidak memfilter iklan dan menjawab plaintext di port 53. Lima pengecekan sebelum Anda mengganti alamat DNS.
- DNS Security
- Security Engineering
-
9 min read
I Checked DMARC on 4,437 Indonesian .id Domains. 47% of go.id Can Still Be Spoofed
A passive DNS scan of 4,437 .id domains: SPF, DMARC, MTA-STS and DNSSEC by second level. 47% of go.id and 74% of all can still be spoofed.
- Security Engineering
- Internet Governance
- Email Security
-
7 min read
I Pointed 3 Static Scanners at an MCP Server That Fails 8 Auth Tests. They Found 0, Including Mine
mcpscan, Semgrep (200 rules) and Bandit gave a broken MCP server and its fixed twin the same report. Five custom Semgrep rules caught 3 of 8 auth bugs.
- MCP Security
- AI Security
-
9 min read
I Read the CI of 300 Popular Repos. 145 Still Run Third-Party Actions by Tag or Branch
I parsed 4,750 workflows from 300 popular repos. 145 of 265 that use third-party actions still reference one by a movable tag or branch.
- Security Engineering
- Engineering Practice
-
8 min read
I Wrote an MCP Server Auth Test Harness for the 2026-07-28 Spec. Here Is What Broke First
Ten tests mapped to the MCP 2026-07-28 authorization spec. A quickstart-style server passed 2. Four fixes and two policy changes got it to 10.
- MCP Security
- Security Engineering
-
10 min read
Let's Encrypt Is Going to 64 Days. I Checked 520 .id Sites. 3 in 4 go.id Sites Still Run Long-Lived Certificates
Let's Encrypt moves to 64-day certificates on 10 Feb 2027. I scanned 520 .id sites: 3 in 4 go.id sites still use certificates valid for 198 days or more.
- Security Engineering
- Internet Governance
-
6 min read
Let's Encrypt Pindah ke 64 Hari. Saya Cek 520 Situs .id: 3 dari 4 Situs go.id Masih Pakai Sertifikat Berumur Panjang
Mulai 10 Feb 2027 sertifikat Let's Encrypt berlaku 64 hari. Saya memindai 520 situs .id: 3 dari 4 situs go.id masih memakai sertifikat berumur panjang.
- Security Engineering
- Internet Governance
-
14 min read
MCP Security: A Practitioner's Guide
MCP security guide: trust boundaries, eight attack classes from tool poisoning to unauthenticated servers, a defense checklist, and how to scan with mcpscan.
- MCP Security
- AI Security
-
9 min read
Python 3.15 Is Out. I Checked the Top 500 PyPI Packages: 97% Install From a Wheel, 12 Still Run Build Scripts
I classified the latest release of the 500 top PyPI packages for Python 3.15. 485 install from a wheel; 12 fall back to running build code, led by PyYAML.
- Engineering Practice
- Security Engineering
-
6 min read
Saya Cek DMARC 4.437 Domain .id. 47% Domain go.id Masih Bisa Dipalsukan
Hasil pemindaian DNS pasif 4.437 domain .id: SPF, DMARC, MTA-STS dan DNSSEC per kategori. Plus langkah perbaikan untuk admin pemda, kampus dan sekolah.
- Security Engineering
- Internet Governance
- Email Security
-
5 min read
Saya Memindai 30 Image Docker Populer. 10 Masih Percaya Root CA yang Sudah Dicabut Chrome
Checklist trust store container dengan data nyata: 30 image Docker dipindai 24 Oktober 2026. Sepertiga masih membawa emSign ECC Root CA - G3.
- Security Engineering
- Deployment
-
3 min read
Server MCP Palsu Menghabiskan Tiga Bulan Membangun Kepercayaan. Tandanya Sudah Ada
Server MCP palsu dari SmartLoader membangun kepercayaan tiga bulan. Sinyal mana yang bisa dipalsukan, mana yang mahal ditiru, dan checklist sebelum install.
- MCP Security
- AI Security
-
8 min read
The eMudhra Fallout, Part 2: I Scanned 30 Docker Images. 10 Still Trust the Root Chrome Is Dropping
Chrome drops emSign ECC Root CA - G3 on October 27. I read the CA bundles in 30 popular Docker images: a third still ship the root, some built after the fix.
- Security Engineering
- Deployment
-
5 min read
UUIDv7 Bagus untuk Primary Key, Buruk sebagai Jam: 4 Tes Sebelum Migrasi
UUIDv7 bagus sebagai primary key database tapi buruk sebagai jam. Empat tes cepat, apa yang dibocorkan timestamp di dalamnya, dan aturan praktis pemakaiannya.
- Security Engineering
- Engineering Practice
-
5 min read
uv check: Cara Type Checking Script Python Single-File dengan uv dan ty
uv check eksperimental dari Astral menambah type checking ty untuk script Python single-file (PEP 723). Coba dalam 15 menit dan lihat seberapa layak dipercaya.
- Engineering Practice
- Open Source
-
6 min read
Kunci CA eMudhra Ternyata Cuma Satu Byte. Container Kamu Masih Percaya?
Kunci intermediate CA eMudhra ternyata cuma satu byte. Browser memblokirnya dalam hitungan jam, tapi trust store container kamu belum tentu. Audit 15 menit.
- Security Engineering
- Deployment
-
6 min read
PHP2Shell: The Patch Is Out. Is It Running in Your Containers?
PHP2Shell patches for CVE-2026-109618 and CVE-2026-109619 are out. Five checks to prove the fix is running in your containers, not just released.
- Security Engineering
- Deployment
-
6 min read
The eMudhra CA Keys Were One Byte Long. Does Your Container Still Trust Them?
Eight eMudhra CA keys were one-byte numbers. Browsers blocked them in hours, but containers may not. A 15-minute trust store audit for Docker, Python, Java.
- Security Engineering
- Deployment
-
6 min read
uv check: Finally, Type Checking for Single-File Python Scripts
Astral's experimental uv check adds ty type checking for PEP 723 single-file Python scripts. Try it in 15 minutes and see how much to trust it so far.
- Engineering Practice
- Open Source
-
6 min read
UUIDv7 Is a Great Database Key and a Bad Clock: 4 Tests Before You Migrate
UUIDv7 is a great database key but a poor clock. Four quick tests, what the embedded timestamp leaks, and a rule of thumb for using UUIDv7 as a primary key.
- Security Engineering
- Engineering Practice
-
7 min read
ICANN87 Hari 5: Saat AI Agent Butuh Nama
Catatan hari kelima ICANN87: SSAC punya work party dampak AI pada DNS, sesi identitas AI agent, sepuluh tahun PTI, dan pelajaran DNSSEC dari para praktisi.
- Internet Governance
- DNS Security
-
7 min read
ICANN87 Hari 4: Dari Aksara Bali sampai Root Server
ICANN87 hari 4: IDN dengan aksara Bali, Jawa, dan Pegon, cara kerja IETF, perjalanan sebuah paket, dan tujuh mitos root server yang ternyata salah.
- Internet Governance
- DNS Security
-
5 min read
ICANN87 Hari 3: Internet yang Diukur, Bukan Ditebak
Catatan ICANN87 hari 3 di Bali: plenary DNS multibahasa, Tech Day ccNSO, dan OCTO DATA dengan DAAR serta ITHI yang bikin kesehatan DNS punya angka.
- Internet Governance
- DNS Security
-
5 min read
Dari Pelatih Keamanan DNS ke Fellow ICANN87: Apa yang Dicari Seorang Insinyur Keamanan Siber di Bali
Fellow PANDI ke ICANN87 di Bali: apa yang dicari insinyur keamanan siber soal DNS abuse, DNSSEC, dan AI agent sebagai klien DNS, plus rencana paper-nya.
- Internet Governance
- DNS Security
-
6 min read
From DNS Security Trainer to ICANN87 Fellow: What a Cybersecurity Engineer Looks for in Bali
A security engineer and PANDI .id Academy trainer heads to ICANN87 in Bali to learn about DNS abuse, DNSSEC and how AI agents use DNS. Here is the plan.
- Internet Governance
- DNS Security
-
6 min read
NetScaler Zero-Day: 3 Checks After Sunday's Patch
Citrix called the NetScaler SAML zero-day CVE-2026-88779 a DoS bug while admins saw code execution. Three checks after the patch, with build numbers.
- Security Operations
- Incident Response
-
6 min read
Wiring a Third-Party Decision Model Into Claude Code, and What I Wouldn't Let It Do by Default
Wiring the Jev decision model into Claude Code: securing the API key first, catching a hallucinated endpoint, and the vendor defaults I kept off.
- AI Security
- Engineering Practice
-
7 min read
9.8 Windows DNS RCE: 3 Triage Checks Before It Repeats SigRed
CVE-2026-69730 is a CVSS 9.8 unauthenticated Windows DNS Server RCE, SigRed's spiritual successor. Build numbers, patch order and a crash tripwire.
- Security Operations
- DNS Security
-
9 min read
Artifactory Is Under Active Attack: 3 Checks in 30 Minutes
Attackers chained two Artifactory CVEs into admin tokens that log as token:anonymous. Three 30-minute checks from the Wiz report for your instance.
- Security Operations
- Incident Response
-
9 min read
Codex Sandbox Escape: 3 Checks Before You Open Another Repo
Two Codex sandbox escapes: a token in shared Node memory and a patch flow that grants write to /. Fixed builds, plus three checks for your own config.
- AI Security
- Security Engineering
-
7 min read
FortiMail Zero-Day: 3 Checks Before You Trust the Patch
A FortiMail zero-day (CVSS 9.8) lets attackers write files and plant an ld.so.preload backdoor. Patching is not enough, so here are three hunt checks.
- Security Operations
- Incident Response
-
10 min read
I Traced 29 CVEs in One MCP Server to 4 Root Causes
29 mcp-atlassian CVEs traced to 4 root causes: DNS rebinding, credential fallback, tool allowlist and file_path. The safe version floor is 0.23.1, not 0.22.0.
- MCP Security
- Security Engineering
-
9 min read
I Traced MaxKB's 10.0 CVE Through the Advisory and the Code
MaxKB's CVSS 10.0 advisory names a missing approval gate, but the v2.10.5-lts fix changed quoting. A read of the advisory, the diff and three checks.
- AI Security
- Security Engineering
-
7 min read
I Traced AgentCore's Package Injection Bug Through 2 Fixes
AWS's Bedrock AgentCore SDK had the same injection bug fixed twice in install_packages(), and both fixes were bypassable. How to check your version.
- AI Security
- Security Engineering
-
7 min read
MCP Costs 72% of Your Context: The Fix Pi Shipped
MCP tool schemas can eat 72% of your context window. How Pi's Codemode and deferred tool loading fix it, plus a minimal registry sketch you can reuse.
- MCP Security
- Engineering Practice
-
10 min read
MCP Servers Had a Rough 48 Hours: 4 Unauthenticated CVEs
Four MCP servers published unauthenticated CVEs in about 35 hours, plus LiteLLM's KEV-listed bypass. Fixed versions and a checklist to audit your MCP servers.
- MCP Security
- Security Engineering
-
9 min read
Mullvad Public DNS Shuts Down Nov 2: 5 Checks I Ran First
Mullvad's public DNS stops on Nov 2. Quad9 filters no ads and answers plaintext on port 53. Five checks to run before you swap the address.
- DNS Security
- Security Engineering
-
7 min read
Read-Only Postgres MCP Servers Fail With 1 SQL Keyword
A crafted COPY ... TO PROGRAM got past AWS's read-only postgres-mcp-server (CVSS 9.6). Four checks for your own SQL validator, and why the role is the boundary.
- MCP Security
- Security Engineering
-
7 min read
5 Skills I Still Learn by Hand While Agents Write Code
About 80% of my added lines come from agents, but the rework was almost all in that 80%. Five skills I still practice by hand, and why.
- Engineering Practice
- AI Security
-
6 min read
The fake browser extension playbook is back. This time it ships as agent skills
Malicious agent skills repeat the fake browser extension playbook: 341 of about 2,632 ClawHub skills were malicious. Five tells to check before you install a skill.
- AI Security
- Security Engineering
-
6 min read
My AI agent said we were under attack. The logs said otherwise.
My AI agent reported a prompt injection attack, but the logs showed none. How a forensic check exposed a false alarm and what I changed in my agent security.
- AI Security
- Security Operations
-
7 min read
No CVE needed: how a GitHub issue hijacked an AI agent
In a demo, one GitHub issue hijacked an AI agent via the GitHub MCP server and leaked private repo data. Why it needs no CVE, and what I changed.
- AI Security
- MCP Security
-
6 min read
The LiteLLM Supply Chain Backdoor: What 3 Hours Means for Your AI Agent Gateway
Two backdoored litellm versions, 1.82.7 and 1.82.8, sat on PyPI on March 24, 2026. What the window means for your AI agent gateway and how to pin builds.
- AI Security
- Security Engineering
-
5 min read
The payment webhook failure I had to inject on purpose
Most webhook tests skip partial database failure. I injected one on purpose in a payment webhook lab and show how a correlation key and review queue recover.
- Integration Reliability
- Security Engineering
-
3 min read
A fake MCP server spent three months earning trust. The tells were there
A fake MCP server from SmartLoader earned trust for three months. See which signals attackers can farm, which are costly to fake, and a pre-install checklist.
- MCP Security
- AI Security
-
6 min read
What Is MCP Security? Common Attacks and How to Scan Your MCP Servers
What is MCP security? Four attacks found testing 11 MCP servers: tool poisoning, command injection, path traversal, planted prompts, and how to scan for them.
- MCP Security
- AI Security
-
3 min read
200,000 exposed MCP servers later, the boring checks still win
Roughly 200,000 vulnerable MCP servers are exposed, mostly through basic config mistakes. Here are five quick checks to run on your own MCP server today.
- MCP Security
- Deployment
-
4 min read
My Scanner Passed Until I Built a Harness That Lied to It on Purpose
I built a harness of synthetic attack fixtures to test my own scanner, mcpscan: 6 of 7 fired, and the miss, a leaked key in a JSON config, taught me the most.
- MCP Security
- Security Engineering
-
3 min read
The NSA published 17 pages on MCP security. Here it is as a checklist you can run today
The NSA's 17-page MCP security guidance as seven checks you can run today: inventory, pinning, tool descriptions, sessions, privileges, egress, logs.
- MCP Security
- AI Security
-
2 min read
Benchmarking My MCP Security Scanner: 14 of 14 Findings, Zero False Positives
MCP security scanner benchmark: 14 of 14 planted findings detected, 0 false positives across 5 fixtures and 24 tools, at a median 82 microseconds per manifest.
- MCP Security
- Security Engineering
-
4 min read
Three PRs to Rowboat in 24 Hours, One of Them Taught Me More Than the Others
Three PRs to Rowboat in 24 hours using Claude Code, plus what auditing ghost-claimed issues taught me about picking issues and reviewing agent diffs.
- Open Source
- Engineering Practice
-
4 min read
A Happy-Path MCP Demo Proves Almost Nothing About Tenant Isolation
A happy-path MCP demo proves little about tenant isolation. What an MCP security preflight checks with static rules and behavioral tests, and what it cannot.
- MCP Security
- AI Security
-
3 min read
What Actually Changes When an MCP Server Leaves Your Laptop
Moving an MCP server behind a gateway: per-request auth, tenant isolation, request-bound approvals and structured logs, plus a lab with 13 runtime tests.
- MCP Security
- Deployment
-
4 min read
A Valid Payment Webhook Can Still Be Dangerous to Process Twice
A valid payment webhook can still corrupt order state when processed twice. Four fixes: raw body HMAC, idempotency, order states and bounded retries.
- Integration Reliability
- Security Engineering
-
9 min read
MCP 2026-07-28 Went Stateless: A Planted Prompt Is a Credential
MCP 2026-07-28 drops sessions and the initialize handshake, so a handle in the context window becomes a credential. Five per-request checks for server authors.
- MCP Security
- AI Security
-
5 min read
The Cursor Allowlist Bypass That Starts With a File Named curl
CVE-2026-22708 is a Cursor allowlist bypass that starts with a file named curl. I built a static shadowing check for it and explain two gaps it cannot close.
- AI Security
- Security Engineering
-
6 min read
My MCP Security Scanner Missed 2026's Worst MCP RCE: Here Is the One-Rule Fix
My mcpscan path traversal rule MCP007 only covered file reads and missed four 2026 MCP CVEs, including a CVSS 9.1 RCE. Here is the write-sink fix.
- MCP Security
- Security Engineering
-
3 min read
Why Quality, DCO Compliance, and Ethics Beat Volume in AI-Driven Engineering
I audited 25+ Tier-1 open-source repos and closed 25 pull requests instead of shipping them. Why quality, DCO compliance and ethics beat volume in AI-driven engineering.
- Open Source
- Engineering Practice
-
3 min read
Auditing Entra ID Accounts and Privileges Without PowerShell Dependencies
Entra ID audit automation: cross-reference accounts against onboarding and offboarding tickets and verify AI Administrator and Security Administrator roles.
- Identity
- Security Automation
-
6 min read
Two ways to reuse a privileged CI token (and my rule only caught one)
A GitHub Actions workflow_run rule that caught only one of two attack shapes, checkout reuse and artifact reuse, and how I split it into MCP019 and MCP020.
- Security Engineering
- Security Automation
-
8 min read
We Planted 10 Vulnerabilities to Test Free Semgrep. It Reported 3.
We planted 10 vulnerabilities in a React and TypeScript app and ran free zero-config Semgrep. It reported 3 of 10. Here is why and how to close the gap.
- Security Engineering
- Engineering Practice
-
3 min read
Automating Enterprise Phishing Triage: SPF, DKIM, DMARC, and Raw Email Extraction
How I automated phishing triage with raw .eml and .msg extraction, SPF, DKIM and DMARC alignment checks, and a KnowBe4 PAB to PhishER webhook pipeline.
- Email Security
- Security Automation
-
5 min read
The MCP attack your code review cannot see
MCP tool poisoning hides instructions in tool descriptions with invisible Unicode that code review misses. mcpscan launched as a static scanner with 12 checks.
- MCP Security
- AI Security
-
5 min read
Your AI agent is the most over-privileged account you own
An AI agent often gets full shell, your API keys and network on day one. Six least-privilege rules to contain prompt injection and shrink the blast radius.
- AI Security
- Identity
-
4 min read
I gave Claude a memory of everything I browse — here's the architecture
How BraveMCP connects a browser extension to Claude Desktop over MCP, with an HTTP bridge, SQLite and ChromaDB hybrid search, and a useful no-LLM fallback.
- MCP Security
- Open Source
-
7 min read
We thought we had location-based MFA. We had something else entirely
Location-based MFA audit in Entra ID Conditional Access: no location policies existed, legacy per-user MFA was doing the prompting, and how to migrate safely.
- Identity
- Security Engineering
-
Medium ↗
I Built an AI-Powered SOC With Tools We Already Owned. Here's the Honest Version.
An honest write-up of building an AI-assisted SOC at Constellar with tools we already owned: what worked, what did not, and what I would do differently.
- Security Operations
- AI Security
-
5 min read
I Shipped My First Cloudflare Worker via GitHub Actions in 47 Minutes (3 Were Wasted on the Wrong API Token)
My first Cloudflare Worker deployed in 47 minutes. Three went to auth error 10000, caused by one wrong API token template. The working Wrangler setup.
- Deployment
- Engineering Practice
-
Medium ↗
Running a Tabletop Exercise Across the Entire Organisation: A Constellar Story
What we learned about pressure, process, and people when the whole company sat down to face a simulated ransomware crisis together.
- Incident Response
- Security Operations
-
Medium ↗
The Hidden Attack Surface: PDF Metadata & Information Leakage
What your PDFs reveal about your infrastructure, and how to stop that reconnaissance in C# and .NET.
- Security Engineering
- Data Privacy
-
Medium ↗
How I Stopped Manually Reading Email Headers and Built a Phishing Analyzer That Does It For Me
A Mini-SOC automation pipeline with Microsoft Power Automate, KnowBe4 PhishER, and Freshservice that cut phishing triage from minutes per ticket to under thirty seconds.
- Email Security
- Security Automation
-
Medium ↗
From Pixels to Privacy: Using OCR + Microsoft Presidio to Redact Sensitive Data in ID Cards
A hands-on guide to detecting and redacting sensitive data like NIK and NRIC from ID cards with Tesseract OCR and Microsoft Presidio, with code and results.
- Data Privacy
- Automation
-
Medium ↗
From Pixels to Pinpoints: My Journey into OSINT Geolocation
Notes from an OSINT geolocation lecture in the SMT program: how image metadata and open map data turn pixels into precise locations, and the ethics that come with it.
- OSINT
- Internet Governance