Why not just use Goose?

Goose is a harness — a general-purpose framework for connecting an AI model to tools, files, and APIs so it can execute tasks. So is Claude Code. So is Aider. They are genuinely useful, and this project is built to run on top of them, not to replace them.

But a harness doesn’t know what you should build. It doesn’t know which dependencies are safe. It doesn’t know what license changed last week, and it doesn’t have an opinion about any of it. You bring the intent, the architecture, the policy, and the verification standards. The harness brings the hands.

The gap that opens

That works for people who already know what they’re doing. A community organizer who needs a tool to document illegal evictions can’t open Goose and say “build me the thing I need” — not because Goose can’t write code, but because nobody told Goose:

• what the thing should actually be,
• which dependencies are trustworthy,
• which ones quietly relicensed to a non-free license last month,
• which architectural patterns protect user data from state actors,
• or which deployment path doesn’t route through surveillance infrastructure.

The harness executes. It doesn’t decide, it doesn’t verify, and it doesn’t refuse.

What this project is

We Can Just Build Things is the layer above the harness. It is the structured intent, the verified dependency catalog, the exclusion policy with enforcement teeth, and the PIE-anchored decision flow that takes someone from “my community has this problem” to “here is the architecture, here are the vetted components, here is the recipe that assembles them under these constraints.”

When an agent finally runs — whether it’s Goose, Claude Code, or anything else — it runs inside a frame of decisions that have already been made, verified, and documented. The skills here aren’t open-ended “do whatever seems right” prompts; they are scoped workflows with policy files read first, stop-and-ask triggers when information is missing, and a three-layer enforcement engine that blocks the commit if something violates the exclusion policy.

The drill and the blueprints

The difference is the difference between giving someone a power drill and giving them a power drill plus the blueprints, the materials list with every board inspected, the building code, and a foreman who won’t let them frame a wall with rotten lumber. Goose is the drill. This is everything else. The harness is interchangeable; the orchestration layer is the product.

Why the distinction is not academic

A harness optimizes for the short head: the demo, the first deploy, the conference talk. In the long tail — month six of production, the relay operator who discovers rate limiting was never added, the organizer who finds out their moderation pipeline was quietly sending user content to a third-party model — the absence of verification compounds into real harm.

This project is built for the person in the long tail: the one whose safety depends on what they build, who needs to make decisions with their eyes open. The operational advisories, the enforcement engine, and the configuration recipes are the mechanisms that move accountability from a storyteller’s good intentions to a stable artifact that can be inspected by the person who bears the risk.