A malicious Claude Code extension in Cursor’s marketplace tried to install a persistent SSH backdoor. This week’s tutorials cover tracing failures across Terraform stacks, troubleshooting GKE traffic…

Read more

LiteLLM’s latest supply-chain attack started with a leaked key that was removed from Git history but never revoked. PHP’s two critical RCEs and Brotli’s decoder flaw make this a week to check runtime…

Read more

AWS reports that its penetration-testing agent exploited 33 of 37 recent CVEs through the intended vulnerability in containerized tests. Anthropic replaced Git after reaching more than 30,000 merges …

Read more

GhostAction is still spreading: an October 8 burst injected fake security-audit workflows into 346 repositories, now hunting credentials in Git history as well as GitHub Actions secrets.
Airbnb's database replay system caught a MySQL upgrade turning a 0.03-second query into a 2.6-second query.
And if your Kubernetes workers have 300 seconds to shut down, a supervisor with a 10-second timeout can make that setting meaningless.

My take: rotating CI secrets is only half the cleanup

Newsworthy stories

• How Airbnb replays real database workloads to catch MySQL performance regressions

• Datadog finds 3,500+ dangerous Kubernetes RBAC bindings to built-in principals

• Can Claude fix itself? Three production incidents from Anthropic’s on-call team

• Testing Jev against Claude and traditional record linkage for identity resolution

Tutorials of the week

• Connect AWS DevOps Agent to OpenSearch logs and traces through MCP

• Build an AI vulnerability harness with evidence-based exploitability checks

• Debug why a 300-second Kubernetes grace period becomes a 10-second shutdown

• Encrypt secrets in Git, CI/CD and Kubernetes with SOPS

• Why multi-tenant best practices can send background work to the wrong tenant

• Make Terraform easier for AI agents by simplifying your infrastructure

• Poison a Go build cache without changing the program’s source

• Explore how PostgreSQL rewrites query plans

• Cut incident detection from 40 seconds to under 10 with OpenTelemetry, Kafka and Flink

• Structure Terraform modules with consistent files, generated docs and pre-commit checks

Podcasts of the week

The Pragmatic Engineer

Building resilient systems with Sam Newman

Stream the latest episode…

Listen now

18 days ago · 80 likes · Gergely Orosz

Projects of the week

Meme of the week

A leaked GitLab issue-by-email address can let someone push code and run CI jobs with its owner’s permissions, even from outside an IP allowlist. This week, you’ll turn Postgres memory management into an interactive explainer and debug a Kubernetes rollout that stalls while the old Pods keep serving traffic. You’ll also see how CedarDB got Doom’s game logic and renderer running in SQL, with multiplayer included.

My take: ask your AI for an explanation you can interact with

Newsworthy stories

• How a leaked GitLab issue-by-email address can push code and bypass IP restrictions

• Why containers are no longer a security boundary

• How an image upload and an SSO flaw opened a path to OpenAI’s internal repositories

• Can your managed Postgres survive a query that exhausts memory?

• Inside vibe-coded credential-harvesting dashboards and live Amazon Bedrock abuse

Tutorials of the week

• Write AWS DevOps Agent Skills from your operational runbooks

• Handle hot shards and isolate large tenants in sharded Postgres

• Run GitHub Actions runners inside Lambda microVMs

• Build on ECS primitives instead of convenience layers

• Run Doom’s game logic and renderer in SQL

• Debug a stuck Kubernetes rollout with a hands-on lab

• Fix EKS pod distribution drift with the Kubernetes descheduler

Videos of the week

Projects of the week

Meme of the week

This week, you’ll learn how to find memory savings in your data structures and limit cache churn when rolling out the change. You’ll also find practical steps for scaling GKE workers from queue depth, turning unused IAM permissions into reviewable pull requests, and allocating AWS commitment discounts based on usage. And if your Postgres query uses the right index but still takes 20 minutes, you’ll see how deleted rows can get in the way.

Watch Luca Berton rebuild a production recovery live

Join us on October 7 (Wed) for a live session with Luca Berton, creator of Ansible Pilot and author of the backup and DR guides for Ansible Automation Platform. An automated change with valid permissions could pass every check and still corrupt production data. Re-running the playbook rebuilds the host, and a snapshot rolls back every good write since. Luca rebuilds the recovery for a real incident live, step by step, into a runbook live attendees keep. Bring an incident that broke production and passed every check, and it might be the one he runs - Save your seat

My take: ask your coding agent for the recovery plan

Newsworthy stories

• Cloudflare saves another 100TB of RAM with math and Rust

• Anthropic threat intelligence report

• How Fairwinds deployed, maintained, and migrated production n8n on EKS

• What changes when AI agents use your software

Tutorials of the week

• How to make Postgres DELETEs not become a performance bottleneck

• Getting started with tokenomics on AWS

• Track when a Kubernetes PVC was last referenced by a Pod

• Autoscale GKE workloads with built-in PromQL metrics

• Automate IAM remediation through your CI/CD pipeline

• Hacking AI customer service agents

• Implement proportional Reserved Instance and Savings Plans chargeback

• Load and serve GGUF models with Transformers

Videos of the week

Projects of the week

Meme of the week

AWS had another rough year in us-east-1, and this week’s retrospective explains why multi-region still doesn’t come easy. Kubernetes v1.37 adds noexec mounts and emptyDir permissions. PostgreSQL 19 lets you pin a query plan the planner abandoned. Tigris got Git running on object storage. There’s a guide to the GitHub Actions caches most workflows skip, a scenario to test whether your AI SOC can trace a GitHub compromise into AWS, and a write-up on letting an agent rewrite a 92M message/day service.

A webhook platform for your AI agents

Most AI agents can’t receive a webhook: they run on a laptop, behind a VPN, or in a sandbox that blocks inbound traffic.

Svix Ingest takes events from Stripe, GitHub, incident.io, or anything else that sends webhooks, and your agent pulls them over an outbound HTTPS connection a few seconds later. No public URL, no tunnel, and nothing lost while it was offline - Try it out

My take: give your coding agent a failing test before asking for a fix

Newsworthy stories

• Finding the cloud waste your billing data misses

• How GLM built its own inference infrastructure

• Why AWS multi-region is still hard, even with AI

• A scenario to evaluate whether your agentic SOC can trace a GitHub workflow compromise into AWS

• Every database has these seven tables

Tutorials of the week

• A practical guide to GitHub Actions caching

• Kubernetes v1.37: hardening container storage

• You don’t need a Mac to develop iOS apps

• Running Claude Code and Codex on a cloud VM

• You can run Git on object storage if you remake packfiles

• Managing Kubernetes node drains with Pod Disruption Budgets

• PostgreSQL row-level security with EF Core and Npgsql

• I got tired of paying JFrog for a secure OpenTofu and Terraform registry, so I built my own

• Plan advice in PostgreSQL 19

• Monitoring the production agent lifecycle with AWS DevOps Agent

• A practical DuckDB and AI workflow for data analysis

Videos of the week

Projects of the week

• OpenDepot is a Kubernetes-native, self-hosted registry for OpenTofu and Terraform modules and providers.

• security-audit-skill turns your coding agent into a security auditor that hands every finding to a fresh agent to disprove before it reaches the report.

• RedAI sends validator agents into a running application to prove or disprove each vulnerability before reporting it.

• litecli is a command-line client for SQLite with autocompletion and syntax highlighting.

• Vex8s generates VEX documents by checking which container CVEs your Kubernetes security contexts already mitigate.

• Cline is a coding agent for your IDE, terminal, and desktop, with a headless mode for CI.

Meme of the week

OpenAI disclosed that, during training, some AI agents left instructions in their handoff notes telling the next agent to hide mistakes from the user.

A team pointed hundreds of Codex and DeepSeek agents at two PaperCut CVEs and compromised 440+ servers at 395 organizations in 48 countries, going from vulnerability research to the first real victim in under four hours.
The same week, Wiz scanned 3,074 internet-facing LiteLLM gateways and found 294 of them still accepting sk-1234, the master key from the docs, with a path from there to the cloud account's IAM credentials.
Add a .git config that makes Claude Code, Codex and Cursor run attacker code on clone, and this is the week the agents started attacking each other's infrastructure.

Faster GitHub Actions, one line to migrate

Avrea is CI infrastructure for GitHub Actions: you keep your workflows and swap runs-on for one of their labels, and jobs run on high clock-speed CPUs with a colocated cache for dependencies, Docker layers, and build artifacts, plus live observability with SSH into the runner when something is flaky.

I tested Komiser’s workflows on their runners, and the cold run went from 5min to 3min, with the Go test suite at 2min instead of 4min.
Free tier is 3,000 build minutes a month, on Linux, macOS and Windows - Try Avrea for free

My take: what Spotify did to cut Claude Code tokens by 90%, and what you can copy

Newsworthy stories

• PaperCut attacker uses hundreds of AI agents to compromise 440+ instances

• Nearly 1 in 10 exposed LiteLLM gateways accepted the example sk-1234 admin key

• Malicious .git configs can make Claude, Codex, Cursor and other AI agents run attacker code

• Portal by Spotify cut my Claude Code token usage by 90%

• 20x the CI traffic without getting slower: how Datadog rebuilt Git serving

• RTK reports token savings, but our cost benchmarks disagree

• I’ve operated petabyte-scale ClickHouse clusters for 5 years

Tutorials of the week

• Implementing chaos engineering in financial payment systems on ECS

• Investigate DMS migration issues with AWS DevOps Agent

• Kubernetes disaster recovery: guidance from three reproducible failure scenarios

• Distributed tracing for CI pipelines without touching a single workflow file

• The Karpenter change that quietly turned on cross-AZ traffic charges

• Breaking Claude Code Opus 5 auto mode: RCE from a website summary

• What SREs should automate, and never automate, with AI

• Automate IAM Identity Center governance with continuous discovery and reporting

• How to handle Terraform CI/CD for multiple environments

Videos of the week

Projects of the week

• walgit hosts git repositories from a single binary in front of an S3 or GCS bucket, with no database, no leader, and no local state that matters.

• Mantis gives coding agents security review skills to find, reproduce, and patch vulnerabilities on their own.

• Cloud in a Bottle deploys and shares web apps on a server you control, with rootless hardened containers and one login for all of them.

• llmfit tells you which open-weight models your CPU, RAM, and GPU can comfortably run, and at what quantization.

• argo9s monitors Argo CD applications in real time from a K9s-style terminal UI.

• dbmask finds the sensitive columns in a SQL database, masks them with deterministic fakes, and verifies the masking actually happened.

• Nightshift wakes your coding agent at night, hands it one small job at a time, and leaves pull requests for the morning.

Meme of the week

Someone spent a month knocking on the root account of more than 150 AWS organizations, and Datadog's writeup is worth reading precisely because every single attempt failed. The interesting part is that the attackers already knew the root email addresses, which makes this a failed campaign built on successful reconnaissance.
Meanwhile Shai-Hulud came back checking 469 places for credentials instead of 189, Nvidia agreed to buy Hugging Face for about $13 billion, and Kubernetes quietly shipped the thing plenty of teams have been faking with cron jobs for years.

One webhook platform for your agents and your customers

Most AI agents can’t receive a webhook: they run on a laptop, behind a VPN, or in a sandbox that blocks inbound traffic.

Svix Ingest takes events from Stripe, GitHub, incident.io, or anything else that sends webhooks, and your agent pulls them over an outbound HTTPS connection a few seconds later. No public URL, no tunnel, and nothing lost while it was offline.

Try Svix for free

My take: three frontier models in three days

Every big lab shipped this week. Claude Fable 5.1, Gemini 3.8 Flash, and GPT-6 Astra.

Newsworthy stories

• Password spraying hit AWS root accounts at 150+ organizations

• Shai-Hulud now scans 469 credential locations, up from 189

• Nvidia agrees to buy Hugging Face for almost $13 billion

• Kubernetes v1.37 can scale workloads to zero with the HorizontalPodAutoscaler

• MCP went stateless, so your AWS deployment can drop sticky sessions

• How GitHub cut Copilot’s inference bill without hurting task quality

• Saving petabytes of cache storage with Zstandard and Pingora

• AI scrapers are now most of kernel.org’s traffic

Tutorials of the week

• Detecting multi-stage attacks on AWS by correlating signals across services

• Terraform observability: drift detection, TF_LOG and plan metrics in practice

• How to handle errors in Go

• AWS S3 bucket security: finding the secrets that live outside Git

• Cutting EKS MTTR with the AWS DevOps Agent and a Kubernetes operator

• Can an AI agent debug Kubernetes? I broke a cluster to find out

• Why pod setup started timing out after 90 seconds

• Which tools do Claude, Codex and Cursor reach for?

• I spent a full day on a “five-minute” Docker migration

• PostgreSQL shared memory visualizer

• Build your own continuous modernization pipeline with AWS Transform custom

• Choosing GCP cost optimization tools that match your environment

Videos of the week

Projects of the week

• Warren runs coding agents as managed infrastructure with isolation, spend limits, and Git delivery instead of terminal sessions.

• Boost compresses the noisy output of the commands your coding agent runs so it keeps the errors and timings without the log spam.

• ZSvirt runs and manages KVM virtual machines from a single web console, including VMware migration.

• Rootprint searches logs and traces directly from object storage, so you can self-host log management instead of renting it.

• pgterm monitors every Postgres database you own from a single terminal dashboard and flags those that need attention.

• ku browses Kubernetes resources, follows logs, and drops you into pods from a keyboard-driven terminal UI.

• srelens combines cluster browsing, logs, exec, and port forwarding into a single local desktop workspace built for engineers and agents alike.

Meme of the week

AWS signed a deal this week to acquire DuckLabs, the ~30-person Amsterdam team behind DuckDB, and the timing is the part worth staring at: it landed nine days after DuckDB previewed v2.0 with a client/server protocol.
Also, Cloudflare shaved 533 bytes off every DNS cache entry and got 100 terabytes of RAM back across its fleet, and I shared a good case for shipping your executables as SQLite databases.

My take: what AWS actually bought when it bought DuckDB

If you haven’t used it, DuckDB is SQLite for analytics. pip install duckdb, point it at a Parquet file in S3, run SQL. No cluster, no Spark job, no ticket to the data platform team. It won because most data is small. AWS’s own post says DuckDB covers the “90%+ of data queries in the world today” that run on data smaller than 1 terabyte.

Four things I’d take from it:

• This is an S3 play, not a Redshift one. The announcement was signed by Andy Warfield, who runs S3.

• The deal landed nine days after DuckDB previewed v2.0 with the Quack protocol, which allows any DuckDB process to serve its databases over the network. Embedded DuckDB is a developer tool. DuckDB that speaks a wire protocol is a query engine you can rent by the second.

• The product writes itself. Quack on top of S3 Tables and Iceberg gives you serverless SQL straight against object storage, no warehouse in the middle. That’s aimed at Databricks and Snowflake, not Redshift.

• AWS didn’t buy the project; it hired everyone who commits to it. No license fight, no fork, no OpenSearch-style backlash. The Foundation keeps the IP, AWS keeps the roadmap.

Newsworthy stories

• How we saved 100 terabytes of memory by optimizing 1.1.1.1’s DNS cache

• On-call is now theatre

• Your executable is a SQLite database

• How Databricks uses AI to accelerate incident investigation

• Stop trying to learn all of Kubernetes at once

Tutorials of the week

• From clickops to governed IaC: CloudFormation drift detection in practice

• Optimizing Kubernetes pod deployments for reliability with topology spread constraints

• How to evaluate LLMs before production

• How we built an AI security analyst on AWS for under $500/month

• Putting models to the secure coding test: plan vs default mode

• Federating clusters for zero-downtime Kubernetes

• How to achieve resource-level isolation in Kubernetes using a multi-environment node pool strategy

• HashiCorp’s tfpolicy made module source a policy decision

• Building safe MCP servers for your PostgreSQL database

• Automating filtered cost and usage report exports with AWS Data Exports

Videos of the week

Projects of the week

• Khaos injects the Kafka failures you normally wait for into a cluster on demand, from consumer lag and hot partitions to rebalances and broker outages.

• Proval reviews your pull requests with an LLM that runs entirely inside your own network, on GitLab, Forgejo, and GitHub alike.

• restoredrill restores your latest Postgres backup into a throwaway container, runs your validation checks against it, and writes the JSON report an auditor actually asks for.

• AWSHound turns your AWS IAM data into a BloodHound graph, so the privilege escalation paths in your account become something you can see.

• cloudflare-github-actions-runner runs your GitHub Actions jobs on ephemeral Cloudflare Containers you size yourself, instead of on GitHub’s runners.

Meme of the week

Cloudflare published the actual economics of AI code review this week: 131,000 reviews a month, $0.98 each, and an 85.7% cache hit rate. Swizec’s counterpoint landed the same week, arguing that most of what those reviewers catch is a deterministic linter rule you could run in seconds for nothing. Meanwhile Cursor rebuilt Git storage on write-ahead logs in S3 to get past the limits consensus-based Spokes ran into, and pgrust rewrote the Postgres query engine with batching, operator fusion and SIMD to make analytics 300x faster.

Newsworthy stories

• Orchestrating AI code review at scale

• Stop burning tokens on code review

• Git at any scale

• Rebuilding Postgres for 300x faster analytics

• On building scalable control planes

• Linux 7.2 brings cache-aware scheduling, faster ext4 and Btrfs

Tutorials of the week

• Massively parallel Postgres backups

• Automating Terraform testing

• Managed AWS dashboards for instant cloud financial visibility

• Building an advanced agentic harness

• PostgreSQL for everything

• How Kubernetes probes work

• A security field guide to AI tooling visibility

• Streamline your GitHub journey with AWS DevOps Agent

Videos of the week

Projects of the week

• Bruno replaces Postman with an offline API client that saves every request as a plain-text file, so collections branch and are reviewed in Git like the rest of your code.

• Munder Difflin runs your terminal coding CLIs as a self-coordinating team of agents on the subscriptions you already pay for.

• secfiles collects wordlists and payload files for penetration testing and bug bounty work.

• OpenViking is a self-evolving context database that unifies agent memory, RAG knowledge, and skills.

• triagebot-action puts every new GitHub issue through reproduce, diagnose, and fix, then ships a preview release and waits for the reporter to confirm before opening the PR.

• Agent Safehouse sandboxes coding agents on macOS so they only reach the files and integrations they actually need.

Meme of the week

Cloudflare pays $0.98 per review for AI to catch things a linter would catch for free. Which one is actually running in your pipeline today, and would you drop the other?

This week’s supply chain attack didn’t drain one company’s secrets, it dumped terabytes of credentials scraped from thousands of repositories into the open. Figma’s answer to that class of noise is a fleet of agents that triage alerts, query security data and write the fix, which cut their resolution time by 71%. One SRE team applied the same instinct to Kubernetes and pulled release validation from 45 minutes down to 2, and if you want a reminder that owning the tech is not the same as winning, the Docker video below is the best bedtime story of the year.

See how PayFit promotes infra and app releases through one pipeline

Join us on September 2 (Wed) to learn how PayFit’s platform team promotes every release the same way, Kubernetes services and Lambda functions alike. One path for app and infrastructure changes, so a release candidate proves itself before the next environment. It’s how PayFit ships payroll for 22,000+ businesses - Save your seat

Newsworthy stories

• Terabytes of credentials leaked in massive supply-chain attack

• How we secure Figma’s internal systems with agents

• How we cut Kubernetes deployment validation from 45 minutes to 2 minutes

• Investigating persistence mechanisms in AWS

• The hidden cost of homegrown IaC

Tutorials of the week

• Track generative AI costs with Amazon Bedrock inference profiles

• The infrastructure testing pyramid

• Pulling multi-gigabyte container images in seconds on Amazon EKS

• A 1,200% AWS cost spike: AI monitoring vs CloudWatch

• Before the first prompt: code execution paths in trusted coding-agent projects

• How to build your infrastructure monitoring in 2026

Videos of the week

Projects of the week

• OpenCodeReview reviews pull requests from the command line and leaves line-level comments.

• sqlfmt formats PostgreSQL queries the way gofmt formats code, aligning clause keywords on a single column.

• GoAccess analyzes web server logs in real time from your terminal or the browser.

• git-knife edits every commit’s message, author, and dates in a table you can click through.

• bsdkrun boots BSD guests, Linux OCI images, and unikernels as microVMs on macOS and Linux, with no daemon running underneath.

• Cloudflare Computer provides agents with a virtual filesystem and swappable execution backends within a Durable Object.

Meme of the week

Terabytes of credentials are now public, so I am curious where you actually stand: do you rotate CI tokens every deploy, every quarter, or only after an incident?

Permiso’s teardown of the OpenClaw ecosystem is the one to read first: agents installing skills from an unvetted marketplace while holding live credentials to email, Slack and everything else they touch. Cloudflare shipped its own answer to that problem this week with Cloudflare OS, an open platform that puts guardrails around what an agent can actually reach. And if you want proof that we are all still improvising, someone wired Claude up as a Postgres query engine and it returned real rows.

Newsworthy stories

• Inside the OpenClaw ecosystem: what happens when AI agents get credentials to everything

• Cloudflare OS: an open platform for agents, apps, and work

• How Figma stays ahead of vulnerabilities with agents

• Turning Claude into Postgres so I can raise a Series A

• Automating multi-cloud cost management at scale

Tutorials of the week

• Accelerate CloudFormation development with the IaC MCP Server

• Everyone records the MySQL audit log. Nobody reads it

• Improve your monthly cloud variance analysis with a weekly FinOps checkpoint

• The multi-account Terraform setup nobody talks about

• The DISTINCT in your COUNT

• Centralize cross-account Amazon ECS telemetry with an ADOT gateway

• What it actually takes to run parallel AI agents

• A security analysis of Amazon S3 Vectors and its use in LLM retrieval pipelines

Videos of the week

Projects of the week

• pgGraph runs graph traversals and shortest-path queries directly on your existing Postgres tables.

• nono sandboxes any AI agent behind least-privilege isolation with no setup.

• glab-tui browses GitLab and GitHub issues, merge requests, pipelines, and runners without leaving your terminal.

• Suzaku hunts threats in AWS CloudTrail logs by applying Sigma detection rules to your API call history.

• Svix handles webhook delivery, retries, and signing so you never have to build that layer yourself.

• Starboard keeps a terminal permanently docked beside the macOS Dock instead of hidden behind a hotkey.

Meme of the week

Attackers turned compromised PHP repositories into a botnet this week, using GitHub Actions runners to scan and exploit cPanel and WHM servers at scale.
It is the clearest sign yet that your CI minutes are somebody else's free infrastructure.
Elsewhere, PlanetScale explains how 768 database servers manage to look like one, and AWS shipped a control plane that finally gives platform teams a policy and cost story for Claude Code.

Learn Rogo’s IaC approach

How do platform engineers keep pace with a growing development team at a $2B AI company? On August 11, Lawrence Aiello, Platform Engineering Lead at Rogo, shares how. He’ll cover the modern IaC platforms Rogo weighed, the criteria that mattered, and how infrastructure that runs without attention frees his team to focus on enabling developers. Take Rogo’s criteria into your own platform decision - Register Now

Newsworthy stories

• Large-scale GitHub Actions abuse powers a distributed cPanel and WHM exploitation campaign

• Making 768 servers look like 1

• Introducing Claude apps gateway for AWS

• Merge conflicts are a process problem, not a Git problem

• Why do Kubernetes experts choose managed Kubernetes-as-a-service?

Tutorials of the week

• Automate CI/CD troubleshooting with AWS DevOps Agent and GitHub

• Safe Terraform auto-apply with conftest

• A green Kubernetes deployment does not mean a healthy application

• I built an AI SRE agent that diagnoses incidents before I open my laptop

• Your pod died of “out of ephemeral storage” but the disk is half empty

• Delightful integration tests in Rust

• A CISO’s guide to agentic AI

Videos of the week

Projects of the week

• tfplantui turns a Terraform plan into a navigable dependency graph in your terminal.

• Widen is a native Postgres GUI for macOS that answers questions in plain English and shows you the SQL before it runs.

• tuicr reviews code in a vim-keybound TUI and exports the result straight to GitHub or GitLab.

• zeedumper dumps the flagz, statusz and configz pages from Kubernetes components through the API server proxy.

• iron-proxy blocks all egress from untrusted workloads by default and swaps sandbox tokens for real credentials at the boundary.

Meme of the week

An attacker opened 37 pull requests against AsyncAPI last week, and eight minutes after the malicious one landed, the CI secrets were gone and four npm packages with 3 million weekly downloads were shipping a three-stage payload. The same week, Socket mapped a network of 222 fake GitHub repos pushing a Go module that was really a Windows malware loader, and Sysdig walked through an Azure tenant that fell in about an hour because permissions live across five systems that don’t talk to each other. The most interesting read, though, might be Rootly explaining why they killed their two-year-old small-PR rule now that 80% of their PRs are written by agents.

Newsworthy stories

• AsyncAPI supply chain compromise via GitHub Actions

• Malicious Go module exposes GitHub malware lure network spanning 222 repositories

• Anatomy of an Azure permission takeover

• Introducing a policy workflow built for Terraform

• Why we got rid of our small-PR rule

• How I migrated my website to Cloudflare and saved $228

• The first year managing an SRE team

Tutorials of the week

• The startup’s Postgres survival guide

• Hunting malware and malicious MCPs in memory on Kubernetes

• How I built my own home-lab: a single-node Kubernetes platform

• Authenticate legitimate AI agent traffic with AWS WAF Bot Control

• FinOps guardrails at provisioning time

• 6 security settings every GitHub maintainer should enable this week

• Multi-cluster databases on Kubernetes: architecture and deployment

• How to back up PostgreSQL in Docker

• From Terraform spaghetti to a production-grade GitOps platform

• To every agent its own database

Videos of the week

Projects of the week

• Cleat gives AI coding agents a per-project Docker sandbox so they can run fully autonomously without touching your host.

• Rusternetes is a ground-up reimplementation of Kubernetes in Rust that already passes 94% of the official conformance suite.

• Destructive Command Guard intercepts destructive shell commands before an AI agent gets to run them.

• OpenGeni is a self-hostable agent runtime with durable sessions, human approvals, and governed credentials.

• frp exposes a local server sitting behind NAT or a firewall to the internet.

• Plannotator is a local browser UI for marking up an agent’s plans and diffs, then sending the feedback straight back to it.

• git-filter-repo rewrites Git history to extract subdirectories, rename paths, or strip secrets.

Meme of the week

Someone opened their AWS console this week to a billing alert projecting $140 billion for the month, on an account that normally spends $2-3 on S3. Underneath the meme-worthy screenshot, the real theme was infrastructure hitting its limits: a kubelet memory leak in Kubernetes 1.36, four recurring failure patterns behind thousands of Postgres outages, and Slack rebuilding its entire EC2 fleet management from scratch. And with half this week’s tutorials about giving AI agents identity, memory, and APIs, the CNCF is already asking whether a pod is even the right place to run one.

AI is hitting your infra faster than you can govern it

AI is reshaping infrastructure, but the teams winning aren't adopting fastest, they're governing best. Spacelift surveyed 406 infrastructure and platform engineering leaders and found 93% have already had an AI-caused incident, while only 30% have a formal AI policy. The report reveals what separates the Pioneers, from platform engineering to the AI-specific signals most teams still aren't tracking.

Download the report to see what the most advanced teams do differently.

Newsworthy stories

• Just got an AWS billing alert projecting my monthly cost at $140B

• Fixing a kubelet memory leak in Kubernetes 1.36

• Inside the unpatched Argo CD RCE that leads to K8s cluster takeover

• Is a pod the right deployment unit for an AI agent?

• How we built Slack’s next-generation EC2 platform

• The four horsemen behind thousands of Postgres outages

• The database is not the data model

Tutorials of the week

• Just use Postgres for job queues

• Cloudflare as a data platform?

• How to encrypt Terraform state files

• Implement a unified memory from scratch via knowledge graphs

• Using Go for mobile apps

• Automating cross-repo documentation with GitHub Agentic Workflows

• Automated incident remediation with AWS DevOps Agent and Kiro CLI

• Detecting full table scans with SQLite

• Solving the identity crisis for AI agents

• Designing APIs for agents

Videos of the week

Projects of the week

• Portero shows every listening port on your machine, which process owns it, and lets you kill it or block the port from a built-in firewall.

• Greysight is an open-source dashboard for Snowflake cost observability that automatically suspends idle warehouses.

• Goshot generates polished code and terminal screenshots with syntax highlighting and automatic secret redaction.

• deja-vu is a memory index for coding agents that searches past session logs so they can recall previous solutions.

• pgsavvy is a terminal UI for PostgreSQL that works like “the lazygit for Postgres.”

• Khoj is a self-hostable AI second brain for semantic search and chat over your own documents.

• LobeHub is a platform for managing a team of AI agents as coordinated, schedulable units of work.

Meme of the week

A single attacker chained together application services, CI/CD pipelines, and secrets stores to compromise an entire AWS environment in 72 hours, and the responders tracking it say human-in-the-loop SIEM triage can’t keep pace with an AI moving that fast. The same week, GitHub’s AI agent was caught handing private repos to anyone who posted a nicely worded issue in a public repo, while an unpatched Argo CD flaw gave attackers full control of Kubernetes clusters. If you were counting on “Trusted Publishing” to keep your supply chain clean, one writeup this week argues that’s the wrong thing to trust.

Platform engineering on autopilot

Mendral is an autonomous AI DevOps engineer that handles the work your engineers shouldn’t be doing manually anymore: supply chain security, flaky CI, slow builds, and anything else specific to your stack - Try it out!

Newsworthy stories

• Lone attacker uses AI to breach AWS cloud environment in 72 hours

• Security flaw in Argo CD can let attackers take over Kubernetes clusters

• I replaced my GitHub runners with Lambda MicroVMs, and maybe you should too

• Why sandboxing your agent is not enough

• You shouldn’t trust Trusted Publishing

• Why AI has made security hard

Tutorials of the week

• Build your own vulnerability harness

• How we tricked GitHub’s AI agent into leaking private repos

• Turn your Amazon CloudWatch alarms into actionable signals

• PostgreSQL as a temporal database

• I built PR preview environments with AWS Lambda MicroVMs and cut staging costs by 78%

• Setting up continuous Terraform drift monitoring

• Experiences with local models for coding

• Three ways to give an AI agent an identity

• Multi-account AWS is about ownership, not accounts

Videos of the week

Projects of the week

• kpt is a package-centric toolchain that treats Kubernetes configuration as data, allowing you to edit and deliver manifests using reusable transformation functions rather than templating.

• Nox is a modular Go framework for attack surface management that bundles 300 built-in modules across 24 phases, taking you from passive OSINT to active vulnerability scanning.

• dotenvx is a secure env manager from the creator of dotenv, encrypting .env files with public-key cryptography.

• AnduinOS 2 is an Ubuntu-based Linux distribution designed to give newcomers a familiar, low-friction path to Linux, packaged almost entirely as shell scripts.

• GeoSQL is a Python skill that turns Claude, Codex, and Copilot into geospatial analytics agents across PostGIS, BigQuery, and Snowflake.

Meme of the week

This week, someone makes the case for keeping your databases off Kubernetes, AWS shares what breaks when you scale to 1 million Lambda functions, and LLMjacking takes an ugly turn: attackers are wiring stolen AI compute into autonomous pentesting agents. Also inside: a production migration off Ingress NGINX, pgvector running at scale on Aurora, and a field guide to corrupting SQLite on purpose.

Newsworthy stories

• Why I haven’t run my databases on Kubernetes

• Lessons learned from scaling to 1 million Lambda functions

• CloudTrail in CloudWatch isn’t very good

• Too many Postgres tables are bad

• How we migrated a production cluster from Ingress NGINX to Gateway API

• Attackers are using stolen AI compute to build offensive agentic tools

Tutorials of the week

• Claude Code for infrastructure as code

• Optimize LLM costs on Amazon Bedrock

• Multi-cluster Kubernetes explained

• Run any Dockerfile on Vercel

• AWS Amplify + Lambda MicroVMs

• Running pgvector in production on Amazon Aurora PostgreSQL

• Kubernetes on a fleet of AWS Lambda MicroVMs

• Blue/green deployments on Kubernetes with Argo Rollouts

• Getting started with Claude loops

• How to corrupt an SQLite database file

Videos of the week

Projects of the week

• terraform-review-agent reviews Terraform PRs with parallel security, cost, and style agents and posts one severity-ranked comment.

• datadog-saist is an AI-native SAST tool in Go that uses LLMs to catch vulnerabilities in Java, Python, and Go.

• ZeroFS serves S3 buckets as POSIX filesystems over NFS, 9P, and NBD, with built-in compression and encryption.

• pgsync syncs Postgres, MySQL, or MariaDB to Elasticsearch or OpenSearch in real time from a JSON document schema.

• ballast is a Kubernetes operator that right-sizes pod CPU and memory requests from real usage with in-place resize.

• tidb is a distributed, MySQL-compatible SQL database with HTAP and native vector search.

• herdr is a Rust terminal multiplexer for running multiple coding agents in parallel workspaces.

Meme of the week

This week’s supply chain news is rough: 32 malicious @redhat-cloud-services npm packages stealing credentials via a preinstall hook, and Mitiga finding spyware and 1,230+ leaked API keys hidden inside AI agent instruction files. On the building side, AWS shipped Lambda MicroVMs for running untrusted and AI-generated code in isolation, Postgres 19 hit beta, and one engineer doubled their test suite by dropping per-test SQL migrations.

Feature deployed. Did it actually work?

A change can pass CI and review and still break checkout the moment it deploys. Cleric follows every agent-written change into production, checks that it did what you meant, and gives you a verdict: it worked, fix this, or roll back - See it work on a real change.

Newsworthy stories

• How to become ridiculously good at Kubernetes

• Malware in AI instruction files, plus 1,230+ leaking API keys

• How to render giant code diffs in the browser

• What’s coming in the Postgres 19 beta

• Cutting SQL migrations from tests doubled the suite speed

• Lambda MicroVMs: what you need to know

• Inside the Red Hat npm “Miasma” credential-stealing campaign

Tutorials of the week

• Running isolated sandboxes with Lambda MicroVMs

• Running AI agents safely inside Kubernetes

• Automating PostgreSQL backups to S3 with Airflow

• Building an incident management workflow

• PostgreSQL 18 performance wins on Aurora and RDS

• The feedback loops behind Kubernetes

• Cloning an ECS environment without rewriting Terraform

• Governing AI assets at scale with MCP Gateway and Registry

Videos of the week

Projects of the week

• no-mistakes is a Go local Git proxy that runs an AI review, test, and lint pipeline in an isolated worktree before pushing, then opens a clean PR.

• Hunk is a terminal diff viewer built for reviewing agent-authored changesets, with inline AI annotations.

• BoxBox is a self-hosted file manager for homelabs that supports chunked and resumable uploads.

  

• dbtrail is a point-in-time recovery tool for MySQL that runs time-travel queries and can undo cascade deletes without locks or restores.

• oh-my-reddit is a terminal Reddit client on Bubble Tea, with live-streaming comments and automatic browser session reuse.

• Orca is an agent orchestrator that runs Claude Code, Codex, and 40+ other CLI agents in parallel within git worktrees from a single desktop app.

Meme of the week

Fable 5 opened a 1,810-line pull request on Tabularis this week. Thirty-seven new tests, all passing, in about half an hour. The line count isn’t the interesting part. It’s that it didn’t just build the feature it was asked for. It refactored the shared connection logic first, as the maintainers said they would have done themselves.

A few days later, Cloudflare pointed a different kind of agent at its own code. A harness that hunts for vulnerabilities, then hands every finding to a second, different model to argue it down. It surfaced 13,841 real bugs across 145 repositories. One model finds, another model doubts, and only what survives reaches a human.

While two teams were teaching AI to build and audit code, someone else was teaching it to sell malware. A Rust clipboard hijacker shipped with roughly 15,500 wallet addresses baked in, fronted by AI-generated YouTube narrators reading polished tutorials to an audience of coordinated fake comments.

Notice the pattern. The same kind of autonomous agent is now writing production code, auditing it, and, in the wrong hands, packaging the attack. The more capable they get, the more of the unglamorous work they can actually own. That’s the bet behind Mendral: an autonomous AI DevOps engineer that takes the jobs your team shouldn’t be doing by hand, like supply chain security, and just handles them.

In the tutorials: how a single prompt injection turned Claude Code’s own GitHub Action into a way to read your API keys, the gameday that proved Datadog’s Postgres failover wasn’t actually safe, and why one NULL can make a NOT IN query quietly return nothing at all.

On video: Postgres branching that clones a database in under a second and scales to zero, plus why slow learning still wins in the AI era.

Then six devtools worth a look, including a Kubernetes desktop client that installs nothing in your cluster, a bastion that fronts SSH, Postgres, and Kubernetes with no client software, and an MCP server that indexes the entire Linux kernel in three minutes.

Platform engineering on autopilot

Mendral is an autonomous AI DevOps engineer that handles the work your engineers shouldn’t be doing manually anymore: supply chain security, flaky CI, slow builds, and anything else specific to your stack - Try it out!

Newsworthy stories

• Why is DuckDB fast?

• Hackers use AI-generated YouTube narrators to promote Crypto malware

• What job interviews taught me about Kubernetes

• Fable 5 opened a 1,810-line, 37-test PR on Tabularis in 30 minutes

• Running local models is good now

• Infrastructure as Code needs software engineering, not more config tooling

• Cursor, GitLab, and Zed agree GitHub is breaking

• The only scalable delete in Postgres is DROP TABLE

• How Cloudflare’s multi-agent harness found 13,841 bugs across 145 repos

Tutorials of the week

• Building high-availability PostgreSQL on Kubernetes

• GitHub Actions security checklist for supply chain attacks

• Cut your AWS Fargate bill by 40%

• Getting started with AKS using Pulumi and C#

• Building a RAG-powered SRE knowledge base with Amazon Bedrock

• Securing CI/CD in an agentic world: Claude Code GitHub action case

• Run your email agent on serverless

• A goroutine leak profile in the Go standard library

• Abusing AI features in SQL Server 2025

• The NULL in your NOT IN

• Operationalizing AWS security: a maturity roadmap

Videos of the week

Projects of the week

• Klustr is a native Go/React Kubernetes desktop client that drives Helm, Argo CD, Flux, and cert-manager from your kubeconfig with no in-cluster components.

• Local Cloud Browser is a native macOS GUI for 28 AWS-compatible services, using Keychain credentials and read-only by default.

• WinPodX runs Windows apps as native Linux windows via FreeRDP RemoteApp, built in Python and Qt6.

• Zvec is Alibaba’s in-process C++ vector database that searches billions of vectors in milliseconds with hybrid retrieval.

• codebase-memory-mcp is an MCP server that builds code knowledge graphs, enabling AI agents to skip re-reading files and index the 28M-line Linux kernel in 3 minutes.

• Warpgate is a transparent Rust bastion and PAM for SSH, HTTPS, Kubernetes, MySQL, and Postgres with no client-side software.

Meme of the week