3. Context phase
Agents, Rules and Skills: the operating system of the AI project.
"AI doesn't understand what you know. It understands what you tell it. And the distance between the two is where most projects are lost."
What it is: The discipline of translating the human understanding produced in Phases 1 and 2 into a system that AI can process. It's not about writing prompts or summarizing what the team knows. It's about designing an operating system composed of three inseparable elements: Agents that orchestrate and execute, Rules that govern and constrain, and Skills that define the concrete work. Together, these three elements are the context.
Who participates: The Context Engineer — the architect of the Agents, Rules, and Skills system. Not a prompt engineer. Not a technical writer. A designer who works simultaneously across three dimensions: defining who does what (Agents), what's allowed and what's not (Rules), and what specific work needs to be done (Skills).
What questions are answered: How do we translate the Problem Statement and Solution Brief into something an AI can act on with precision? What roles does the AI need? What rules are non-negotiable? How do we decompose the solution into self-contained units of work? How do we ensure that nothing gets lost in translation?
What it produces: A complete context system: a project-context.md with the Rules that govern everything, Agent definitions with identity, responsibilities and limits, a PRD, an Architecture Document, and Story Files that are the Skills each Agent will execute. Plus a Decision Log that records every evolution.
When it's complete: When the outputs produced by Agents executing Skills under Rules are recognized by the stakeholders from Phases 1 and 2 as faithful expressions of the Problem Statement and Solution Brief. Not when the documents look good. When the results are right.
Context Engineering is systems design. You don't produce documents — you produce an operating system that AI will execute. The difference between a good prompt and a good context system is the same as between giving an instruction and building an organization. Instructions are forgotten. Systems operate.
Why
The gap between what a team understands and what AI can process is the context. Context Phase is the discipline of crossing it.
Context vs prompt
A prompt is a point-in-time instruction. Context is an information architecture composed of Agents, Rules, and Skills. Confusing the two is the most common mistake.
Who
The Context Engineer is the architect of the Agents, Rules, and Skills system. Not a prompt engineer — a system designer who works across three dimensions simultaneously.
Rules
Rules are the norms that apply to the entire project. They are what turns a collection of independent Agents and Skills into a coherent system.
Agents
Agents are specialized AI profiles — not generic instances. Each one has a defined role, clear limits, and a protocol for when information is insufficient.
Skills
Skills are the 'what gets done concretely.' Self-contained work units with all the context an Agent needs to complete them without ambiguity.
→ Phase 4
When the Gate Review is complete, the team has a system of Agents, Rules, and Skills. The construction phase doesn't start with a briefing — it starts with context.