The Groundwork Guide.
The complete, open program: six layers, built in order, across three phases. Fifteen documents. Then one Operator. Nothing gated, nothing held back.
How this program runs
Groundwork is not a promise. It is a folder of files. You build fifteen plain-text documents in a fixed order, and those documents become the foundation any AI can stand on. The order is not optional, because each layer depends on the one below it.
Phase 1 is required first: Identity, Knowledge, and Governance. Who you are, what the system may trust, and what it may do. These three layers alone improve output quality immediately — before you automate anything.
Phase 2 activates one Operator. A single scoped agent. Not autonomous. Not a swarm. It drafts, cites authority, and stops when it should.
Phase 3 makes it durable: Continuity, Measurement, and Operating Rhythm. What survives failure, how you know it is working, and how it is maintained so it does not decay.
The rule: do not automate chaos. Most teams skip the foundation and go straight to agents. The result is generic output, invented policy, and systems that cannot tell doctrine from a draft. If your identity is unclear, your knowledge is unclassified, and your rules are unwritten, an agent will amplify the mess — faster than you can catch it.
The three phases
Foundation install
Identity, Knowledge, and Governance. Eleven documents. Output quality and decision boundaries improve immediately, before a single agent exists.
First Operator live
One scoped agent, running real work inside the foundation. Lock the scope, load the package, run real Draft & Wait work, and correct the documents from actual misses.
Built to last
Continuity, Measurement, and Operating Rhythm. Four documents that make the foundation survive time, people, and drift.
The map of the guide
Nine parts. Read Foundations first, build the layers in order, activate the Operator after Phase 1, then score yourself. Every part is open.
Foundations
Understand the machine you are feeding: the model, the agents, the instructions, and the data — in plain language.
Outcome: the mental model. You know why context beats prompts and why the build order is not optional.
Layer 1: Identity
Who you are and how you are allowed to behave. Charter, brand voice, and decision principles.
Outcome: a context file any AI can load, so output sounds like you and decisions follow your rules.
Layer 2: Knowledge
What the system is allowed to trust and use. Authority Classification, the Context Vault, priorities, and constraints.
Outcome: an AI that cites your files at the right authority level, not the internet's guesses.
Layer 3: Governance
What the system is allowed to do. Auto-Execute, Draft & Wait, Never Automate — plus escalation rules, allowed tools, and agent scope.
Outcome: written rules a human signs off on, so automation stays inside boundaries you set.
The Operator
One scoped agent, activated only after Phase 1 exists. The context package, the prompt, and the rules for adding roles later.
Outcome: a competent internal operator that drafts, cites authority, and stops when it should.
Layer 4: Continuity
What survives time, people, and failure. Succession rules, override authority, and working memory discipline.
Outcome: a foundation that survives the founder being gone for ninety days.
Layer 5: Measurement
Whether the foundation is actually working. The signals, the review questions, and the correction loop.
Outcome: you know when it is working — and when it is drifting — before drift becomes damage.
Layer 6: Operating Rhythm
How the foundation is maintained so it does not decay. Ownership, review cadence, and the update protocol.
Outcome: a living foundation, not a one-time build.
The Readiness Scorecard
A weighted self-assessment across all six layers. Runs entirely in your browser.
Outcome: a score, your weakest layer named, and a clear answer to “what do I fix first?”
What you have when it is done
- Fifteen plain-text documents that define who you are, what your system trusts, and what it may do. Yours, on your machine, readable by any AI.
- An AI that sounds like you. Output stays on-voice without re-prompting, because the voice is written down.
- Authority the system must cite. Five levels, 0–4, so a rumor and a contract are never treated as equal.
- One working Operator. A scoped agent running real Draft & Wait work with a human signing off where it matters.
- Platform independence. Markdown files work with Claude, ChatGPT, Gemini, or any future AI. No lock-in. Ever.
- A foundation that outlives you being busy, absent, or gone. Succession, override authority, and a review rhythm that keeps it honest.
What this costs
Nothing. The whole program is on this site and in the Field Guide PDF, free and ungated. No email required. No paywall. No locked chapters. If you want a second set of eyes, book the free Audit — sixty to ninety minutes mapping your current state and your biggest gaps. Questions: adam@adamabdalla.com.