ContextQB

ContextQB teaches people how to build with AI agents without surrendering architectural control.

AI can write code. Someone still has to think.

The project combines a methodology, a learning library, a course platform, an MCP server, and a repo-health toolchain for people using agentic IDEs to build real software. Its focus is not better prompting in the shallow sense. It teaches the missing discipline: how to give agents enough structure, context, standards, and review loops that the work remains coherent as the project grows.

Problem

Agentic coding has moved the bottleneck from typing to thinking.

Tools like Cursor, Claude Desktop, Windsurf, and other MCP-aware environments can produce working code quickly. But speed is not the same as direction. Without project context, architectural boundaries, naming discipline, state ownership, documentation, and review habits, an AI-built codebase can become tangled faster than a human can understand what happened.

Most agentic-development guidance teaches prompts. ContextQB teaches the operating system around the prompt: the maps, principles, playbooks, audits, manifests, and feedback loops that make an agent useful instead of merely productive.

Concept

ContextQB is an architecture system for AI-native builders.

It has four connected layers:

Methodology — principles, guides, playbooks, audits, prompts, examples, and briefings that teach how to plan, build, inspect, harden, and maintain agent-built software.

MCP server — a tool interface that exposes the ContextQB library directly inside agentic development environments, so an agent can retrieve the right principle, playbook, audit, or prompt during the work instead of relying on whatever context happens to be in the chat.

context.qb.yaml — a boot manifest format for describing a repository to an agent: what the project is, where the apps live, what routes deploy where, what conventions hold, and which architectural decisions matter.

contextqb check — a drift detector that keeps the boot manifest honest by comparing it against the repository structure, deployment configuration, and architectural decision records. A stale map is worse than no map; the check turns staleness into something the build can catch.

The larger idea is simple: in agentic development, documentation is architecture. What is not written down cannot reliably govern the agent.

What it offers

ContextQB gives builders a practical system for steering AI work:

• principles for architecture, naming, state ownership, modularity, documentation, security, and maintenance
• playbooks for repo setup, feature planning, agent instructions, architecture review, refactoring, security review, and incident response
• audit templates that turn vague “review this” requests into structured inspection documents
• prompts and briefings designed to produce decision-grade output instead of conversational drift
• an MCP server that makes the library available to agents inside the tools where the work happens
• a boot manifest and drift detector that keep the agent’s map aligned with the real repository

The audience is everyday builders, founders, operators, first-time software makers, and experienced developers who want a shared language for agentic work.

Why it matters

The early AI-coding conversation has been too agent-centered.

The real leverage is on the human side of the desk: the structure of the repo, the clarity of the instructions, the quality of the project map, the discipline of the review process, and the standards that persist across sessions.

ContextQB is a bet that agentic development needs more than smarter models. It needs better operators.

It gives non-developers a way to think architecturally, gives developers a way to standardize agent collaboration, and gives agents a clearer substrate to work against. It is not anti-speed. It is speed with steering.

Status

ContextQB is active. The public methodology site is live, including guides, principles, playbooks, audits, prompts, examples, and briefings. The MCP server exposes those resources to agentic development tools. The context.qb.yaml boot manifest pattern and contextqb check drift detector are in use as part of the broader system.

Course and member/community layers are in development, including shared insight tooling for patterns across agentic projects.