Strategy-as-Code Repo Template
A git repository scaffold for encoding strategy as code — /intent, /charters, /policies, /rewards, and doctrine.md — with a pull-request workflow, a Risk-Twin CI gate, and a branch-protection rule that stops agents from amending their own charters.
├ intent/ ├ charters/ ├ policies/ ├ rewards/ └ doctrine.md
What it does
"In software, code that cannot compile does not run. In an AI-Born enterprise, strategy that cannot compile does not execute — it creates a vacuum that agents fill with their best interpretation of what you probably meant." Chapter 5 defines Strategy-as-Code as "the discipline of encoding strategic intent precisely enough that agents can act on it, test it, and report on it — and precisely enough that violations surface in real time rather than in a post-mortem." This template is that discipline, scaffolded.
It ships the exact repository the book describes. At Adaptic, strategy lives "in a git-backed strategy repository — /intent, /charters, /policies, /rewards, and a doctrine.md that states what the company believes. A merged pull request to that repository rewrites live agent behavior within seconds. The strategy is not a deck that describes the company; it is the source the company is compiled from." So the template gives you those four directories, the doctrine file, charter schemas, and a PR workflow where a single commit shapes thousands of agent interactions — Chapter 7's "asymmetric leverage."
The charter schema is the book's seven elements, not a freeform prompt: identity and classification, role definition, decision rights, forbidden actions, objective function, escalation policy, and performance monitoring. The template validates against Strategy-as-Code's three named properties — precise (a satisfaction objective without a margin floor is "a blank check"), version-controlled (every change tracked, attributed, reversible), and testable (run against historical or simulated data before deploy). Chapter 7's Meridian example is the spec made vivid: a CPO added "a 36-month customer lifetime value constraint and a hard cap," a change that "took four minutes to commit and three to deploy" and "protected $400 million in transaction volume."
Two gates make it safe. First, a Risk-Twin CI check: every proposed change to /policies or /rewards runs against the simulation before merge (developed in Chapter 6). Second — and this is an extension consistent with the book rather than a verbatim rule — branch protection enforces a no-self-amendment principle: agents cannot merge to /charters, /intent, /rewards, or doctrine.md. The book never lets an agent "freelance into decisions that belong to a human," insists "the charter must precede the agent," and locates amendment authority with humans ("who can authorize a charter amendment"). The template encodes that as CODEOWNERS plus required human review.
Who it's for: founders, Guardians, and platform engineers standing up the governance substrate for a Machine Core — and anyone tired of strategy that lives in a deck nothing can execute.
Figure: The repository this template scaffolds — the source the company is compiled from.
The source the company is compiled from. A merged PR rewrites live agent behavior within seconds — a single commit shapes thousands of interactions, asymmetric leverage that amplifies both wisdom and mistakes.
strategy/doctrine.md # what the company believes 🔒intent/ # IPRE Intent outputs 🔒sales.mdsupport.mdcharters/ # 7-element Agent Charters 🔒vp-sales-agent.charter.yamldeal-desk-specialist.charter.yamlvp-support-agent.charter.yamlpolicies/ # governing policy versionssales.policy.yamlsupport.policy.yamlrewards/ # objective functions / reward weights 🔒vp-sales-agent.rewards.yamlvp-support-agent.rewards.yamlCODEOWNERS # human ownership of protected paths 🔒.ci/validate.yaml # three-properties checksrisk-twin-gate.yaml # simulation gate on /policies & /rewards
🔒 = protected path · agents cannot merge here (no-self-amendment design principle — a human CODEOWNER must approve). No /guardrails directory: guardrails live inside charters.
Risk-Twin endpoint is unset — the CI gate is scaffolded as a required-but-stubbed check that fails closed (stays red) until configured, rather than silently passing.
- 1. CI validate — three-properties checks. Fail = no merge.
- 2. Risk-Twin gate — normal + adversarial scenarios (ch6). Fails closed (endpoint unset).
- 3. CODEOWNERS human review on protected paths.
- 4. No-self-amendment — agent commits to charters/intent/rewards/doctrine rejected.
- 5. Merge rewrites live behavior within seconds. Pre-authorized categories take the fast-path.
Strategy as Code
Agent Charters
IPRE Pipeline
Risk Twins
Risk Twin Scenario Planner
IPRE Pipeline Designer
Alignment Debt Tracker
Agent Charter Builder
- Book 1, Chapter 5 — "The Operating System: How the Machine Core Runs" (Strategy-as-Code; the git-backed strategy repository; the seven-element Agent Charter; the three properties; charters precede agents).
- Book 1, Chapter 7 — "Strategy as Code" (the commit as commitment; asymmetric leverage; Meridian's four-minute amendment; pre-authorized amendment categories).