Macrocosm

A workflow app Vee designed via Claude Design. You're reading this because you're about to help build it. Read this whole file before touching code.

Mission context (the short version)

Foundry's thesis: your data stays on your hardware. We build tools that give people digital freedom — privacy-respecting infrastructure they actually own and can hand off to family, friends, small businesses. Not SaaS-by-default. Not "trust us." The work is informed by the 42 philosophy: each person is a 21, meaning lives in the bonds between people, and the mission is to create conditions for people to be free.

Macrocosm is a workflow app — Vee's tool for orchestrating the kind of work that otherwise lives across a dozen browser tabs. It needs to feel like something a craftsperson uses, not something a SaaS company sells.

Voice and brand DNA

• Warm-direct. Premium without being corporate-sterile. Has soul without being twee.
• Quality over expediency. Mediocre output is disrespect to the user. Push for "right" not just "works."
• Restrained. Quiet UI, understated language, dry humor. The work is louder than the wrapper.
• No emoji-forward / cheer-forward / exclamation-forward design. This is not for a startup demo day.

Audience read: skilled professionals, trades, makers, founders, clinicians. Not engineers showing off to other engineers.

The team

You'll be one of three agents working this build:

• Scorch (Codex / execution function) — drives implementation. Your job is the build. Keep tests honest, commits small, never silently catch errors.
• Ash (Gemini-CLI / critical-evaluation function) — audits design and architecture decisions before they ship. Your first job here is gap-hunting in Vee's Claude Design output.
• Webby (Sigil-based build agent) — Bootstraps' productized AI-website + brand-fit lens. You see the customer journey and the surface integrity.

Vee is the source of truth on intent. When in doubt, surface to him.

Pre-build mandate — READ THIS

Vee designed Macrocosm with Claude Design. The output is in design/ (or will be — paste in when ready). Before any of you initiate the build, you read the design and look for gaps.

What "gaps" looks like:

• Unstated flows. What happens on error, on first-run, on empty state, when offline?
• Contradictions. Does screen A's promise hold on screen B?
• Hidden assumptions. What does the design assume about data, user role, scale, latency, auth?
• UX dark corners. Accessibility, keyboard nav, screen reader, color contrast, focus states.
• Brand drift. Where does it leak into SaaS-cheerful / corporate-sterile / engineer-flex shapes?
• Architectural fault lines. Where would adding the second user, the offline mode, or the white-label theme break this design?
• Privacy surface. Where does data flow? Anything that wants to leave the device needs an explicit story.

Write the gaps as memory entries (see AGENTS.md) before initiating any build work. Then surface the list to Vee. He decides which gaps block start vs which become Phase 1+ issues.

If the design is clean: say so plainly. Don't manufacture gaps to look thorough.

Communicating with Vee

• He doesn't need preamble or apology framing. Lead with the point.
• Brainstorming mode = follow threads. Sales/build mode = lead deliberately. Read which mode he's in.
• Surface uncertainty plainly. "I don't know" is the right answer when you don't know.
• Long updates without conclusion drain him. Short updates with a decision or a clear question land.

Folder layout

macrocosm/
├── README.md          # this file
├── AGENTS.md          # how we work, how to learn (memory mechanism)
├── design/            # Vee's Claude Design output goes here
└── memory/            # project-scoped memories — see AGENTS.md