This week on Ship It Weekly, Brian looks at three different versions of the same problem: systems are getting faster, but human attention is still the bottleneck.

We start with curl shutting down their bug bounty program after getting flooded with low-quality “AI slop” reports. It’s not a “security vs maintainers” story, it’s an incentives and signal-to-noise story. When the cost to generate reports goes to zero, you basically DoS the people doing triage.

Next, AWS improved RDS Blue/Green Deployments to cut writer switchover downtime to typically ~5 seconds or less (single-region). That’s a big deal, but “fast switchover” doesn’t automatically mean “safe upgrade.” Your connection pooling, retries, and app behavior still decide whether it’s a blip or a cascade.

Third, Amazon ECR added cross-repository layer sharing. Sounds small, but if you’ve got a lot of repos and you’re constantly rebuilding/pushing the same base layers, this can reduce storage duplication and speed up pushes in real fleets.

Lightning round covers a practical Kubernetes clientcmd write-up, a solid “robust Helm charts” post, a traceroute-on-steroids style tool, and Docker Kanvas as another signal that vendors are trying to make “local-to-cloud” workflows feel less painful.

We wrap with Honeycomb’s interim report on their extended EU outage, and the part that always hits hardest in long incidents: managing engineer energy and coordination over multiple days is a first-class reliability concern.

Links from this episode

curl bug bounties shutdown https://github.com/curl/curl/pull/20312

RDS Blue/Green faster switchover https://aws.amazon.com/about-aws/whats-new/2026/01/amazon-rds-blue-green-deployments-reduces-downtime/

ECR cross-repo layer sharing https://aws.amazon.com/about-aws/whats-new/2026/01/amazon-ecr-cross-repository-layer-sharing/

Kubernetes clientcmd apiserver access https://kubernetes.io/blog/2026/01/19/clientcmd-apiserver-access/

Building robust Helm charts https://www.willmunn.xyz/devops/helm/kubernetes/2026/01/17/building-robust-helm-charts.html

ttl tool https://github.com/lance0/ttl

Docker Kanvas (InfoQ) https://www.infoq.com/news/2026/01/docker-kanvas-cloud-deployment/

Honeycomb EU interim report https://status.honeycomb.io/incidents/pjzh0mtqw3vt

SRE Weekly issue #504 https://sreweekly.com/sre-weekly-issue-504/

More episodes + details: https://shipitweekly.fm

This week on Ship It Weekly, the theme is simple: the automation layer has become a control plane, and that changes how you should think about risk.

We start with n8n’s latest critical vulnerability, CVE-2026-21877. This one is different from the unauth “Ni8mare” issue we covered in Episode 12. It’s authenticated RCE, which means the real question isn’t only “is it internet exposed,” it’s who can log in, who can create or modify workflows, and what those workflows can reach. Takeaway: treat workflow automation tools like CI systems. They run code, they hold credentials, and they can pivot into real infrastructure.

Next is GitHub’s new fine-grained permission for artifact metadata. Small change, big least-privilege implications for Actions workflows. It’s also a good forcing function to clean up permission sprawl across repos.

Third is AWS’s DevOps Agent story, and the best part is that it’s not hype. It’s a real look at what it takes to operationalize agents: evaluation, observability into tool calls/decisions, and control loops with brakes and approvals. Prototype is cheap. Reliability is the work.

Lightning round: GitHub secret scanning changes that can quietly impact governance, a punchy Claude Code “guardrails aren’t guaranteed” reminder, Block’s Goose as another example of agent workflows getting productized, and OpenCode as an “agent runner” pattern worth watching if you’re experimenting locally.

Links

n8n CVE-2026-21877 (authenticated RCE) https://thehackernews.com/2026/01/n8n-warns-of-cvss-100-rce-vulnerability.html?m=1

Episode 12 (n8n “Ni8mare” / CVE-2026-21858) https://www.tellerstech.com/ship-it-weekly/n8n-critical-cve-cve-2026-21858-aws-gpu-capacity-blocks-price-hike-netflix-temporal/

GitHub: fine-grained permission for artifact metadata (GA) https://github.blog/changelog/2026-01-13-new-fine-grained-permission-for-artifact-metadata-is-now-generally-available/

GitHub secret scanning: extended metadata auto-enabled (Feb 18) https://github.blog/changelog/2026-01-15-secret-scanning-extended-metadata-to-be-automatically-enabled-for-certain-repositories/

Claude Code issue thread (Bedrock guardrails gap) https://github.com/anthropics/claude-code/issues/17118

Block Goose (tutorial + sessions/context) https://block.github.io/goose/docs/tutorials/rpi https://block.github.io/goose/docs/guides/sessions/smart-context-management

OpenCode https://opencode.ai

More episodes + details: https://shipitweekly.fm

This is a guest conversation episode of Ship It Weekly (separate from the weekly news recaps).

In this Ship It: Conversations episode I talk with Gracious James Eluvathingal about TARS, his “human-in-the-loop” fixer bot wired into CI/CD.

We get into why he built it in the first place, how he stitches together n8n, GitHub, SSH, and guardrailed commands, and what it actually looks like when an AI agent helps with incident response without being allowed to nuke prod. We also dig into rollback phases, where humans stay in the loop, and why validating every LLM output before acting on it is the single most important guardrail.

If you’re curious about AI agents in pipelines but hate the idea of a fully autonomous “ops bot,” this one is very much about the middle ground: segmenting workflows, limiting blast radius, and using agents to reduce toil instead of replace engineers.

Gracious also walks through where he’d like to take TARS next (Terraform, infra-level decisions, more tools) and gives some solid advice for teams who want to experiment with agents in CI/CD without starting with “let’s give it root and see what happens.”

Links from the episode:

Gracious on LinkedIn: https://www.linkedin.com/in/gracious-james-eluvathingal

TARS overview post: https://www.linkedin.com/posts/gracious-james-eluvathingal_aiagents-devops-automation-activity-7391064503892987904-psQ4

If you found this useful, share it with the person on your team who’s poking at AI automation and worrying about guardrails.

More information on our website: https://shipitweekly.fm