Was reading about the malicious Claude code extension that show up in Cusor marketplace. The pary that got me was it trying to add an SSH key and open a reverse tunnel l, from something that just look like another editor extension

We talk a lot abt checking npm package and dependencies, but IDE extension feal like one of those thing people install once and forget abt it. Especially with auto updates enbled

Those things make me wonder how much access we are giving these extension without really thinking abt it. Are people actually restricting extension on work machines or is it mostly up to each dev? Curious if anyone team changed anything after this 🤔

Source:https://cursor.com/blog/claude-code-extension-security-notic

I manage a lot of repos, that are changed by multiple devs and every few months it turns into a mess: hundreds of merged branches nobody deleted, PRs that went quiet a year ago, repos nobody has touched since a project ended. GitHub only lets me deal with it one repo at a time, so mostly I just... don't.

Curious how others handle this:

- Do you run a script (gh CLI, an Action), use a tool, or just ignore it?

- What's the most annoying part?

- What would you never trust a tool to do automatically?

Especially interested in hearing from people managing 30+ repos or a team org.

I'm about three-quarters of the way migrating my application infrastructure from ArgoCD to Flux, and man, is it difficult. I had to use both Claude and ChatGPT pretty extensively to parse the absolute mountain of error logs I was getting and correct the pile of bugs that I had inadvertently introduced while setting it up the first time around.

I'm too far into this process to back out, but just in case I finish this process and then realize that I actually didn't want to migrate, which platform is actually used more commonly in the industry, so I should actually spend more energy learning it?

Note, even though I do mention that I used two AI models quite a bit, I only used them for parsing logs and deciding what to do. All of the code is actually handwritten.

I've been voluntold to analyse our Datadog costs because our bills keep going up, and we're paying extra for new features if we go over the committed spend. It's a bit mismanaged across the org. Naturally, I need billing data to do the analysis (and do some MI outside of Datadog).

These endpoints all return 403 Forbidden:

• /api/v2/usage/historical_cost
• /api/v2/usage/estimated_cost
• /api/v2/cost_by_tag/monthly_cost_attribution

I'm an Admin with all relevant billing and usage permissions. Other usage endpoints work fine, and I can see usage data in the UI.

We buy Datadog through a reseller, and I suspect our organisation is a child org under the reseller's parent org. I think that explains why I can't access historical cost data through the API.

Our reseller has gone quiet for quite some time. I've contacted Datadog Sales and Support (I have no idea why we have a direct connection via a shared Slack channel even though we're not direct customers), but they've basically ghosted me.

We're paying for the service, our bills are increasing, and I'm trying to understand where the money goes. Why is accessing our own cost data such a nightmare?

Has anyone dealt with anything similar?

Hi everyone,

What are some quality conferences in Europe worth visiting in 2027?

As the title says, a new colleague recently joined our team, and he uses AI for everything. Not a single Slack message of his comes from his fingertips unfiltered.

The real pain point has been PR reviews. Our team is small, so reviews are hard to come by. I expected that having another teammate would make things smoother since we could share the review load, but his reviews don't add value, and they actually slow down our development cycle more than before.

I like to break my changes into small, digestible chunks that are easy to review. The problem is that he runs Opus 5.5 on xhigh against every change, and it nitpicks things that aren't even in the PR. It could be a two-line documentation change, and he'll comment that some existing code elsewhere in the same file, outside the scope of the change, needs to be modified. He then marks the pull request as "Changes requested" and doesn't re-review it for days on end.

How would you go about discussing this kindly with him?

30K~ DAU, I have full ownership over an infra under AWS/multiple on premise servers.

we use ECS, terraform, instead of ansible I use a deploy.sh that i give to my boxes with terraform, no graceful deploys. we have our own orchestrator (K8S like) for stateful containers that we have to spin up and down very often (median 1 min of uptime per container) with unsafe code coming from AI/users.

I took the place of someone who left that company and explained me nothing, we are 2 devs, nobody understands how this entire infra works expect me. have been adding observability a lot using cloudwatch, multiple alerts so I page myself at the right times, it's stable but we need to scale to at least 5x (an objective the hierarchy gave me). it currently costs 3.5k/month in aws/on premise servers. btw it's not a classical rest api where you run stateless apis and that's it.

I just discovered how redis worked under a distributed system like this, it's quite good :) I have been a postgresql guy but redis has been a saver for our threat model.

I often read here that you should put everything : EKS + K8S, ansible, terraform, make HA, grafana etc.. but since I manage this entire infra I have been resilient to all of this. If I refactor everything right now they have nobody to do the backend (which is also what I do here) -> "startup" role where one guy has multiple roles.

for the terraform it's all AI, I have been doing a lot of modifications and have created a terraform plan under ci cd so I can review myself, I removed a lot of "destruction" like the database, i versioned the s3, versioned the terraform state, put it to a private s3 etc...

there have been big big security issues (xss, csrf, cookie stealing, untrusted code that owns the entire box, shared passwords, untrusted s3 location so you could overwrite other users...) that were there before I got here

nothing rly interesting maybe for you guys in this experience but I have learned a lot, it has been a crescendo experience where i'm getting to understand more and more which tools to put there because I am the one who owns it 100%, and have to also do features (backend) so not much time either.

Hey !

I guess most of us have a lot of admin accesses to tools like AWS, k8s clusters, Datadog, GitHub etc.

How do you contain your agents to read-only or curated permissions? Without restraining your own accesses.

I do have some sort of hook as a guard + a setup on pi.dev that copy my config and extract the read-only accesses only to do so, but it's never completely opaque.

I was checking out how to build a live operational dashboard for a logistics app and found out about composed vs. converged HTAP architectures.
I’ve always been warned never to run heavy analytical aggregations on the primary transactional DB because the contention spikes your transaction latency. But a guide I was reading explained how HTAP systems are handling this now. A “converged” system runs both through the same engine (which sounds risky for resource contention), while a “composed” system actually separates the compute engines and keeps them in sync through shared storage or replication, so each side stays optimised for its own workload.
Apparently, you’re supposed to measure p99 transaction latency while the analytical queries are running, not separately.
If you are running live analytics on operational data, how are you keeping the two from interfering with each other?

we were on buildjet until it shut down in march, been back on stock github runners since and the builds are noticeably slower. team of about 25, linux x64 mostly with a bit of arm

blacksmith, depot, namespace and avrea all look the same on the pricing pages so i cant pick, and every comparison i find is written by one of them lol. also seeing warpbuild and runs-on mentioned

if you picked one and stuck with it, what made you choose it? or did you try a couple and bounce off one

I have an interview tomorrow for a DevOps Engineer position (I have 5+ YOE). The recruiter explicitly told me the interview will be a deep-dive through my CV.

I’m in a massive bind regarding my current/most recent role and need some urgent advice on how to frame it.

The Situation:

• Early this year: I was the lead engineering resource tasked with getting a major product production-ready. The literal day it was ready, a massive reorg hit. The product was stripped from me, and most of its target end-users were laid off.
• Immediately following: I was pivoted to head up another large project for a different internal product. I busted my ass, completed it successfully, and my manager was impressed. Then... another reorg hit. This product was also taken away.
• The past 5+ months: My team was merged with another to form a new SRE team for a completely separate product. The scope was entirely undefined. In fact, one colleague's sole task for the last 4 months was just figuring out what our responsibilities are.

Because of this reorg limbo, I have had zero actual engineering work to do for months. I’ve repeatedly met with my manager and skip-level to voice concerns, but I just get generic corporate hand-waving about a "clear roadmap ahead."

The Dilemma for Tomorrow:
If they deep-dive into what I am doing right now, I see three options, and they all feel terrible:

1. The Brutal Truth: Admit I’ve been sitting on my hands doing nothing due to corporate dysfunction. (Worried this looks toxic or makes me look unmotivated).
2. The Total Lie: Pretend I’m still actively working on those two successful deployment projects and that they were never taken away. (Worried because my poor recall will make it obvious I haven't touched the code recently, and I'll have to fabricate metrics, current challenges, and future roadmaps).
3. The Middle Ground: Pretend I just transitioned off those projects a few weeks ago (similar issue as option 2).

How do I handle this in a technical deep-dive without sounding like a liar or a slacker? Has anyone successfully spun "reorg purgatory" to a hiring manager?

Thanks in advance!

TL;DR: 5+ YOE DevOps engineer with a CV deep-dive interview tomorrow. This year, I built and completed two major projects, but back-to-back corporate reorgs stripped them away. I've spent the last 4 months in team-merger purgatory doing absolutely nothing. Do I tell the truth, lie and pretend I'm still on the projects, or find a middle ground without looking like a slacker?

Every time someone posts about a leaked API key, the top advice is the same. Put it in .env, add .env to .gitignore, done.

After the OpenCode breach and the LiteLLM mess this month I went back through the leaked key stories people posted this year. Found 10, with bills from $2 to about $87K. Not one was "I pushed my .env to GitHub".

Where the keys actually were:

> In the JS bundle. A VITE_ variable in a gitignored .env, repo scanner said clean, then the build shipped it to every visitor. One scan of 978 AI-built apps found 37 doing this.

> In an agent config file. LiteLLM's master key sat in opencode.json, got force pushed over but never revoked, and was used 136 days later.

> In someone else's tool. Keys saved in OpenCode's console got drained, one person lost $5,200 of company credits.

> In a WordPress backup plugin. One root AWS key reused across sites, tested on Aug 23, used 56 days later for about $8K of Bedrock.

> In old Google keys that quietly gained generative AI access. That's the $87K one.

> In nobody's leak at all. Someone got into an Anthropic console through email login and created 3 new keys, $5,473 in a few hours.

And then last week Railway got hit. Attackers decrypted customer environment variables straight from the platform, and Railway told everyone to rotate every key. Gitignored, never pushed, stolen anyway.

What actually helped was boring:

> Hard spending caps. The one person who had one lost a few minutes of usage and that was it.

> An alert isn't a cap. The $87K team's budget alert fired at $550, five hours in.

> Revoke, don't clean up. Force pushing doesn't un-leak anything.

> Delete keys you don't use, and know where every key lives (Infisical, Vault, whatever you already have).

> Don't count on a refund. The $87K bill got $55K off, they still owe about $32K.

I'm not saying stop using .env locally. I'm saying "it's gitignored" isn't the end of the conversation.

On 21 October 2026, Google released patched versions of its Brotli compression library for CVE-2026-111497, a vulnerability nicknamed Node2Shell. It is an out-of-bounds write in the Brotli decoder. Node and many Python stacks bundle their own copy of Brotli, so patching libbrotli on the host doesn't fix them.

I made an interactive, hands-on lab exercise on this vulnerability. The goal is to give engineers a general overview of the attack vector and the response. You'll exploit a service as an attacker, then switch roles to the on-call engineer: upgrade the runtimes, rebuild and redeploy the images, verify the fix, and rotate the deploy token the attacker read from memory.

I did my best to highlight all the details, but if you notice any issues or inconsistencies, please let me know!

https://ransomleak.com/exercises/http2shell-rce/

Hey guys,

I'm preparing to get the CCA and OCTA cert to become golden kubestronuat, i see that the kode cloud cuurses are 20 hrs each, is it really necessary to go with that courses? also i saw that kyverno kode cloud course is so deprecated, is this true? i need to get them fast because cks will expire and i dont want to do it again xD

Thank you so much.

Got my undergrad in EE, doing CI/CD and DevOps work now that I love and want to grow in. ideal graduate programs would be:

1. Takes job experience as waivers for prerequisites I may not have (EE math transfers but I lack CS classic education)

2. 8 week courses (give or take)

3. If CS isn’t the best MS to pursue I’d love to hear alternatives

Thanks to all in advance!

You add alerts because you want to catch problems early.

CPU spikes. Memory usage. Failed requests. Slow queries. Queue delays.

Eventually, the team starts receiving notifications throughout the day.

Some are important. Most aren't.

And that's when a dangerous habit develops: people start assuming the next alert probably doesn't matter either.

I'm curious how teams deal with this.

Have you ever deliberately removed a large number of alerts to make monitoring more useful?

How do you decide which alerts deserve to interrupt an engineer and which should just be visible on a dashboard?