Content Agent Operating System

An open repository for building controlled, evidence-driven content operations.

Draft freely. Validate deterministically. Approve explicitly. Publish deliberately.

The Content Agent Operating System is a repository and reference implementation for teams that want content automation to remain useful without becoming opaque or unattended. It models every job as a small, controlled chain: an accountable manager routes the work, specialists return narrow artifacts, deterministic checks validate state, a named owner approves the exact final artifact, and a scoped adapter carries out one auditable action.

Why this repository exists

Most content automation discussions stop at prompting or model selection. This project starts with the operational system around the model: source ownership, role separation, hard validation gates, explicit approval, idempotency, and an audit record. The goal is not artificial autonomy; it is traceable completion.

Repository map

Area	What it contains	Start here
client/	The React interface that turns the operating model into an editorial reference experience.	client/src/pages/Home.tsx
docs/	The source-grounded architecture guide, control loop, roles, rollout sequence, and economics notes.	docs/architecture.md
media branch	Repository-hosted visual assets used by this project’s documentation and presentation.	View the media collection
ideas.md	The design system and brand rules behind the “Operative Ledger” presentation.	Read the design direction

The work graph

flowchart LR
  A[Owned brief and source assets] --> B[Manager agent]
  B --> C[Specialist workers]
  C --> D[Deterministic validation]
  D --> E[Named human approval]
  E --> F[Scoped official publishing adapter]
  F --> G[Idempotent receipt and audit event]

Loading

Authority at a glance

Role	Responsible for	Must not do
Manager	Route the next constrained action and select allowlisted workers.	Self-approve, invent tools, or publish from informal instructions.
Specialists	Return research, writing, edit-plan, and QA artifacts.	Alter policy or advance unvalidated work.
Deterministic controls	Verify schema, state, rights, budget, duplicate uploads, and evidence.	Make editorial or authority judgments.
Approver + scoped adapter	Approve one artifact and make one defined platform action.	Silently retry or approve a changed artifact.

A controlled job, step by step

1. Load the job with the brief, rights, policy version, budget, and current state.
2. Produce one manager next action naming the target state, worker/tool, required validations, and risks.
3. Validate the policy and schema before any worker runs.
4. Create or validate one bounded artifact.
5. Require named approval for the exact final artifact hash.
6. Record the action with an idempotency key, receipt, and audit event.

30-day adoption sequence

Window	Focus	Evidence gate
Days 01–03	Foundation	Source ownership, consent, rights, brand voice, and a versioned playbook.
Days 04–07	Drafting	Structured intake, source cards, drafts, variants, and a human review packet.
Days 08–14	Controls	State machine, schema checks, idempotency, budgets, and pass/fail gates.
Days 15–21	Production	Technical QA, subtitles, approval packets, and artifact-hash match.
Days 22–30	Publication	One official adapter, receipt, labeled analytics, and a pause procedure.

Run locally

pnpm install
pnpm dev
pnpm check
pnpm build

Scope

This repository is grounded in the supplied Content Agent Operating System PDF. The directional token figures in the reference materials are illustrative and do not include rendering, storage, data transfer, platform APIs, human review, or incident handling. The project is an operating-model resource, not legal, platform, or rights advice.