Tighten access controls and isolation in a homelab
A user asked Claude Opus to harden their homelab's access controls; it locked itself out and then refused to continue, so they switched to OpenAI, which finished the job. ·
Done withClaude
- What they did
- The user asked Claude Opus to tighten access controls and isolation in their homelab. Opus produced a plausible-looking design and was told to go ahead. Partway through it locked itself out of something and asked the user to run root commands. The user, away from the computer, told it to sort it out or use the root login it had. They then switched to OpenAI.
- How it went
- Opus refused to continue, returning a message citing legal/AUP. OpenAI's model, called Sol, picked up where Claude left off and finished the project, though it seemed a little less smart.
- Worth knowing
- The agent may lock itself out mid-change, so stay reachable or give it a recovery path; here a refusal after that lockout ended the Claude session.
Try it yourself with Claude
I want to harden access controls and network isolation in my homelab, which has [list of hosts, VLANs, services, and how I connect to them]. First, audit the current setup and write a step-by-step plan, with a rollback step and a way to confirm I can still reach each machine after every change. Do not apply any change that could cut off SSH, remote access, or your own connection until you have shown me the exact command and I have approved it, and stop when the plan is done and each change is verified.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: HN · Nov 9, 2026
Related use cases
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Survey AI product names with a swarm of search agents
The author had a swarm of Claude agents run 4,654 searches to catalog AI bot names and analyze naming trends by gender, industry and year.
Grade student submissions from Canvas against a rubric
An instructor's pipeline pulls student files from Canvas, has Claude score each per criterion with comments, and posts grades back after human review.
Turn tickets into merged pull requests
Pulls tasks from GitHub Issues or Linear, then runs Claude Code or Codex in isolated Kubernetes pods to carry each one through to a PR.
Scaffold a new service end to end with agents
Cursor agents create Terraform, k8s manifests and CI, validate them and open a draft PR; one cleanup run wrongly deleted Terraform files.
Run Claude Code in an isolated VM that opens PRs
The author runs Claude Code in a restricted VM with secrets served from Infisical. It can only create issues and PRs on Forgejo, and merges are manual before CI deploys.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.