AI Control Plane

Purpose

The AI Control Plane is the operating layer for Prox OS work that begins as founder intent, product judgment, architecture judgment, or business/growth ideas and then becomes reviewable engineering work.

It is manual opt-in governance. Normal Direct Implementation Mode work should not start here.

It does not replace AGENTS.md, docs/development/ai/*, docs/development/rfcs/*, or docs/development/adr/*. It sits above them as the coordination layer for:

Mission briefs.
ADR and RFC routing.
Dated task graphs.
Hard-scoped worker prompts.
Branch-based execution.
QA and quality gates.
Integration reports.
Release plan updates.
Run logs.

Operating Flow

Opt-in governance request
  -> AI Inbox / Obsidian / Airtable raw log
  -> Architect AI mission brief
  -> RFC or ADR when needed
  -> dated task graph
  -> scoped worker prompts
  -> branch-based execution
  -> QA / quality gates
  -> integration report
  -> release plan update
  -> roadmap status update

Naming Rule

Every new feature, fix, RFC, ADR, task graph, release plan, integration report, and agent run should use a date-prefixed filename:

YYYY-MM-DD-short-kebab-title.md

Examples:

2026-05-19-pricing-app-promotion-ui.md
2026-05-19-hola-app-url-open-flow.md
2026-05-19-ai-control-plane-initialization.md

When a unique identifier is useful, add it inside the document:

Prompt ID: P20260519-TASK-001
Run ID: R20260519-001
Decision ID: ADR-2026-05-19-001
RFC ID: RFC-2026-05-19-001

Current vs Future

Current monorepo phase:

This directory is the AI-readable control plane.
Worker prompts still run manually in Codex, ChatGPT, Cursor, Claude Code, or terminal agents.
docs:sync mirrors these docs into the Fumadocs app.

Future multi-repo phase:

A dedicated control-plane repo may hold projects.yaml, agents.yaml, release plans, task graphs, ADR/RFC indexes, prompt templates, and run logs.
Architect AI reads the control plane and assigns scoped work to the correct repo.
Integrator AI reviews branches/PRs across repos before a human owner decides merge and deploy.

Directories

Directory	Purpose
`agent-scopes/`	Role boundaries for Architect, Docs, UI, Shell, API, QA, Integrator, Business, Growth, Content, and Ops agents.
`prompt-templates/`	Copy-paste prompts for planning, RFC/ADR creation, worker tasks, QA, integration, business, and growth.
`task-graphs/`	Dated task graphs that split work before execution.
`rfcs/`	Control-plane RFC drafts and examples. Canonical implementation RFCs still belong in `docs/development/rfcs/` unless this directory is explicitly adopted.
`adrs/`	Control-plane ADR examples and workflow notes. Accepted architecture ADRs still belong in `docs/platform/architecture/decisions/`.
`release-plans/`	Version and slice closure plans.
`integration-reports/`	Read-only merge and integration reviews.
`runs/`	Agent run logs for traceability.

Guardrails

Do not let a large prompt become a large patch by default.
Do not route ordinary UI incubation, mock apps, docs cleanup, copywriting, or Storybook work through the control plane by default.
Do not let multiple agents edit the same ownership area at the same time.
Do not skip required reading, allowed paths, forbidden paths, acceptance checks, or validation commands.
Do not treat future ideas as current implementation scope.
Do not add secrets, tokens, payment logic, production migrations, or deployment changes through a generic worker task.