Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
Lobster Trap: OpenClaw in Containers from Local to K8s and Back — Sally Ann O'Malley, Red Hat
TL;DR
Sally Ann O'Malley’s core pitch is simple: run OpenClaw in containers, not natively — she argues containers give you reproducibility, cleaner machines, better isolation, easier backup via volumes, and a straight path from laptop to Kubernetes or OpenShift.
She treats the ‘OpenClaw is a security nightmare’ critique as a challenge, not a dealbreaker — after 10 years at Red Hat working on containers, Linux security, Kubernetes, and OpenShift, she says the whole point is proving that even risky apps can be run securely with the right sandboxing and secret handling.
Her personal demo makes the tooling concrete: a ‘forever claw’ with sub-agents for astrology and Boston Bruins updates — Joy handles Jyotish astrology, Bruno does daily Bruins briefings, and she backs the whole setup up nightly because she actually relies on it.
Secrets management is the practical centerpiece of her setup — she uses Podman secrets mounted into containers, then OpenClaw’s own secret-ref feature on top, creating what she calls a “pointer to a secret ref to the outside secret” so API keys stay out of logs and configs.
The local-to-Kubernetes story is the real enterprise angle — O'Malley imagines company-standard OpenClaw baselines for new hires, complete with approved MCP servers, auth, team-specific skills, and storage, then personalized on top instead of every employee hand-assembling their own stack.
She says AI is already changing how engineers work, not replacing them — citing a friend at Nvidia using OpenClaw for model evals with about 10 engineers on Kubernetes, she frames the productivity gain as freeing teams from tedious coding so they can do more creative work.
The Breakdown
From Red Hat container veteran to accidental OpenClaw evangelist
Sally opens with the arc: seven years deep in containers, Linux security, Kubernetes, and OpenShift at Red Hat, then a move into emerging tech where AI took over everything about three years ago. She jokes that AI work initially felt like “another chatbot, more Python, more markdown” until a staycation and a viral “Malt book” moment sent her to GitHub to try OpenClaw.
“It’s a security nightmare” — and why that didn’t scare her off
When coworkers warned her not to put OpenClaw on a work laptop, she basically took that as a dare. Her response is pure Red Hat energy: if they can’t take an application and run it securely in containers, what have they been doing for the last decade? That tension sets up the talk’s real thesis: containerization is how you make experimental AI tools usable in the real world.
Her ‘forever claw’: astrology, Bruins, and a very personal use case
She gives the talk a heartbeat by introducing “Shubra,” her forever claw, plus two sub-agents: Joy for Jyotish astrology and Bruno for Bruins playoff briefings. It’s goofy on purpose, but it lands the point — these aren’t abstract demos for her; they’re tools she actually keeps alive, backed up nightly, and wants to trust.
Why containers fit agents so well
Her list is straightforward but persuasive: containers are reproducible, portable across x86, Mac, laptop, and Kubernetes, and naturally sandboxed because access has to be explicit. She also likes that you can mount an entire agent directory — tools, skills, MCP servers — so startup is consistent, and keep runtime state in Podman volumes or Kubernetes PVCs for clean backup and recovery.
Secrets, secret refs, and the double layer of indirection
This is where she gets opinionated. She runs everything with Podman and leans hard on Podman secrets, mounting API keys into containers instead of spraying env vars around; then she stacks OpenClaw secret refs on top of that. It’s not perfect, she says, but it gives her peace of mind and keeps credentials out of logs while preserving the same pattern in Kubernetes secrets.
Her bigger vision: OpenClaw everywhere, from laptop to enterprise Kubernetes
O'Malley says AI workloads are going to run everywhere, and OpenClaws communicating across environments is where this is headed. She imagines workplace setups where new hires get a curated baseline OpenClaw with company-approved MCP servers, auth, Google Drive access, and team-specific skills — then personalize from there, instead of piecing together somebody else’s repo.
The Nvidia anecdote and her blunt take on coding with AI
At PyTorchCon, she says a friend at Nvidia told her they’re using OpenClaw for model evals with about 10 engineers, each running OpenClaw in Kubernetes, and that it works so well it feels like six engineers’ worth of leverage. She immediately pushes back on the job-loss panic: the point isn’t fewer people, it’s less tedious work. Her most provocative line is that AI is now “1,000 times better than me at writing code,” and she says she hasn’t really written code in months.
The live demo: opinionated installer, Podman command, and OpenClaw in seconds
The final stretch is a fast demo of her personal installer: pick a pod name, port, secret mappings, model providers like OpenRouter and Anthropic, optionally wire in OpenTelemetry and Jaeger, and even enable OpenClaw’s SSH sandbox. She spins up a Podman container named Joe in basically no time, then flips to examples running in kind and OpenShift too — ending with a very on-brand, rushed “run OpenClaw in containers… try it.”
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
