Feature-First Delivery Setup

Feature-First Delivery Setup

This is Bloom's Feature Brief workflow packaged as a reusable product, not a one-off local prompt. One Marketplace install carries five jellies (the AI agents inside a crew) and the linked operating knowledge every agent loads while it works.

Who's in it

• 🧭 Compass grounds scope in product intent, existing Features, wikis, manifests, repository evidence, and cited research.
• 📐 Blueprint writes the implementation plan and holds the human approval gate before delivery work fans out.
• 🖼️ Canvas makes interaction choices inspectable with designs, previews, mockups, screenshots, and pictured Decisions.
• 🛠️ Foundry builds only from an approved plan, keeps child work under the parent Feature, and updates the product narrative as reality changes.
• 🔬 Proof validates the whole Feature in the correct environment and records where a human can see it, what passed, and what remains uncertain.

Shared knowledge graph

Five bound categories teach the same through-line to every stage: the Feature Brief as product contract, visible Decision discipline, connected documentation, delivery evidence, and research grounded in Bloom's vectors, graph relationships, source records, and citations. The pages explain which truth belongs in a Feature, artifact, Decision, wiki, manifest, or handoff—so documentation builds on itself instead of becoming another flat list.

What install does

Creates a NEW Feature-First Delivery Crew and adds its knowledge categories, pages, relationships, and jelly bindings to your project in one action. Existing crews and edited pages are not overwritten. Installing again creates another crew, so tune the installed copy instead of using reinstall as an upgrade mechanism.

Set up on any Windows computer

1. Install this setup from Marketplace into the target Bloom organization.
2. Open the new crew, review its model/signoff choices, and assign it to the project and voyage you want it to operate.
3. On each computer that will run local agents, download the Bloom Command Station from Bloom and run the installer it gives you (Bloom-Desktop-Setup.exe). The station is the desktop app that runs agents on that machine; it updates itself from then on.
4. Open Settings in the station, paste an Organization API key into the Bloom card, and press Connect. The key is the only thing to enter: Bloom answers with the organization it belongs to, and every call the station makes is scoped to that organization.
5. Add the project's repository on the Repositories screen, then open Crew and choose Pull from Bloom: the station brings this organization's crew down to that machine and runs its work through whichever agent tool (Claude Code, Codex, Cursor, GitHub Copilot) is installed there.

Marketplace distributes the portable setup; the Command Station is the local runtime that connects a developer's codebase and agent tools to it. That separation lets the same setup run on many machines without baking API keys, paths, repository state, or organization ids into the bundle.

What to customize

• Replace the starter research and documentation pages with your actual product mission, vocabulary, architecture, standards, and business goals.
• Keep plan and QA signoff on for high-consequence work; relax them only where your team has an explicit faster path.
• Bind additional project-specific knowledge to the stable jelly keys compass, blueprint, canvas, foundry, and proof.
• Pair it with Ship's Scribe when you also want a standing daily product story across the project and voyage manifests.