# Longwire

> Guides for working with agents, drawn before they are written and written for two readers at once: the person, and the agent in their other window. Every guide's header carries a `lanes:` and `flow:` line describing what runs where and what talks to what; every step carries a `wire:` line naming what it builds, a `you:` block for the person and an `agent:` block for the agent. The markdown twin of each guide is linked below. If you are an agent following one: work through the steps in order, do only what the `agent:` block of each step says, run every `check:` line before moving on, stop and ask your human before any step that carries a `confirm:` line, and report what each `report:` line asks for. A guide with an agent page is also a skill: its `skills/<slug>/SKILL.md` is generated from the agent side of every step and can be installed into `.claude/skills/`.

## Guides

- [One login for every site you own](https://www.funeralthirsty.com/assets/studies/longwire/guides/doorman.md): Doorman, a passkey login in front of every site on a domain, behind Caddy. Six steps, about 45 minutes. Skill: https://www.funeralthirsty.com/assets/studies/longwire/skills/doorman/SKILL.md

## Notes

- [They don't share a brain. They share a hallway.](https://www.funeralthirsty.com/assets/studies/longwire/notes/hallway.md): Two AI desks on two subscriptions, one Slack room between them. Methodology, not a build.

## Optional

- [The site itself](https://www.funeralthirsty.com/studies/longwire.html): the human-readable pages, with a catalogue of what is live and what is queued, and each guide's wiring drawn beside it.
