# Workflow Design Bible

**A constitution + document-system generator for autonomous, agent-run projects.**

`workflow-design-bible` is a reusable **meta system prompt** (and an installable agent skill). Point it at a brand-new project idea — a content channel, an ebook press, an SEO tool-site, a web product, a casual game, anything meant to *run itself* — and it interviews you, then scaffolds a complete, opinionated workflow:

- a **thin `CLAUDE.md` boot router** that stays resident every turn but holds almost nothing; - a **document system** under `documentation/` — a fixed set of named single-source docs (`CONSTITUTION.md`, `WORKFLOW.md`, `ROLES.md`, `ARCHITECTURE.md`, `IDENTITY.md`, `SOUL.md`, `MEMORY.md`, `NEXT_SESSION.md`, …) loaded **once per session, not every turn** (and some, like the architecture baseline, only when a task reaches for them); - a **session lifecycle** — `/start-session` → work → `/finalize-session` — so each session boots from where the last one left off and ends by improving itself, plus a periodic `/slim-docs` diet so the document system gets *lighter* over time, not heavier; - a **growing identity and soul**, so the agent becomes a work *partner* that gets more itself over time, not a stateless tool.

It encodes one architecture that has proven itself on real autonomous projects:

> **The main agent is a CEO, not a worker.** It orchestrates; sub-agents do the > work; skills document capabilities; a deterministic CLI executes them. Every > session ends by reflecting on itself and freezing "decided on the fly" into > "frozen into a function."

— distilled into **eleven design philosophies** every generated project inherits.

---

## Why this exists

LLM agents are great at *doing a task* and bad at *running a project*. Without a durable structure they re-decide the same things every session, bloat their own context, lose track of what's already built, forget who they are, and never improve their own process.

This Bible bake