Skip to content

Agent skill

vpcatlas ships an Agent Skill: a SKILL.md with reference files next to it, in the open Agent Skills format. It teaches a coding agent or assistant how to run vpcatlas safely and how to read its outputs, so the agent answers from real data instead of guessing. The skill lives in the repository at skills/vpcatlas.

Install it into an agent

  1. Copy the skills/vpcatlas directory into the directory your agent loads skills from, or symlink it there. Each agent documents its own location, so check its docs for the exact path.
  2. If your agent cannot load skills from a folder, point it at skills/vpcatlas/SKILL.md instead and let it read the reference files from there.
  3. Make sure vpcatlas 0.5.0 or newer is on the PATH of the agent, and that the agent has AWS credentials that can read the account.

What it covers

  • The network. Accounts, regions, VPCs, subnets, route tables, gateways (internet, NAT, Transit Gateway, VPN, Direct Connect), peering, VPC endpoints, load balancers, firewalls, DNS and shared resources (RAM), plus the cross-account links.
  • Diagrams and inventories. draw.io or Lucid diagrams, and an inventory as JSON, Markdown or HTML.
  • Answers from real data. Questions about VPCs, subnets, routing, peering, Transit Gateway or endpoints, answered from a scan of the account instead of from memory.
  • Read-only. It writes a graph JSON, an inventory and diagrams, and never changes anything in AWS.

Keep it honest

The skill is written against the CLI source in src/vpcatlas/cli.py and the pages under docs/cli/. A change to a command, a flag or an exit code is not finished until the skill matches it.

An agent must never be given write credentials for this tool. Read access to the account is all it needs.

The files are in the skills/vpcatlas folder.