cd /news/ai-agents/yeah-i-gave-it-access-to-my-email · home › topics › ai-agents › article
[ARTICLE · art-173951] src=forum.level1techs.com ↗ pub= topic=ai-agents verified=true sentiment=· neutral

Yeah, I gave it access to my email

A Level1Techs forum post titled "Yeah, I gave it access to my email" presents an llm.txt prompt that instructs an AI coding agent to build a Proxmox homelab based on the author's earlier configuration post, asking the agent to confirm which coding harness (Pi, OpenCode, Claude Code, Codex, or another) and which services to deploy before proceeding. The prompt directs the agent to check the node's version, resources, storage, bridges, and existing guests read-only, map each requested service to a new guest or reused service with installer links from community-scripts.org, and wait for approval before installing one service at a time and verifying each with a real request. It also bars pasting passwords, tokens, or private keys into chat, requires stopping and reporting on failure, and asks for access instructions, test results, and undo steps at the end.

recommendation

Scope the agent with an llm.txt that requires harness and service confirmation, read-only node inspection, a per-service plan with installer links, explicit approval, one-at-a-time installs verified by a real request, and documented undo steps.

more in playbook: Designing agent systems
read1 min views3 publishedNov 21, 2026

llm.txt

Help me build a Proxmox homelab similar to this Level1Techs guide: https://forum.level1techs.com/t/yeah-i-gave-it-access-to-my-email/260970 . Use the author’s earlier config post as the starting point: https://forum.level1techs.com/t/253398 . Where the posts differ, follow the newer guide. Adapt it to my hardware; don’t copy the author’s network settings. 1. Ask me: “Which coding harness do you want to use (Pi, OpenCode, Claude Code, Codex, or another)? Which services from the guide do you want?” 2. Ask which Proxmox node to use and how to authenticate. Use an existing SSH alias or a scoped credential from my secret store. If I need to sign in, have me do it locally. Never ask me to paste passwords, tokens, or private keys into chat. 3. Check the node version, available resources, storage, bridges, and existing guests read-only. Show a short plan mapping each requested service to a new guest or an existing service to reuse. Link its installer from https://community-scripts.org/scripts . Flag anything without a matching script, including hardware-specific setup. Wait for my approval. 4. Inspect each selected installer and its dependencies before running it. Install one service at a time, keep it private, and verify it with a real request before moving on. Don’t overwrite guests or change host networking. If a step fails, stop and report what changed. Finish with access instructions, test results, and how to undo the new installs.
── more in #ai-agents 4 stories · sorted by recency
── more on @level1techs 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/yeah-i-gave-it-acces…] indexed:0 read:1min 2026-11-21 · —