Skip to main content

3. Context phase

Agents, Rules and Skills: the operating system of the AI project.

This is the operational guide for executing Context Phase — the phase where the team translates human understanding into a system that AI can process faithfully. While the methodology section explains the why and the conceptual foundations (what Agents, Rules, and Skills are and why they matter), this guide covers the how: the concrete process to follow, the artifacts to produce, the criteria to meet before moving forward, and the failure patterns to watch for.

5
Structured steps
From Rules to Validation
6
Artifacts
Transferred to Phase 4
9
Gate Review criteria
All must pass to advance

The Context Phase workflow​

Context Phase translates human understanding into a system that AI can process faithfully, through five steps in strict sequence. Rules come first because everything else depends on them. Agents are designed before Skills because Skills need to know who will execute them. Strategic documents come before decomposition because Skills are derived from them. Validation is last because it tests the complete system, not individual pieces.

1. Establish Rules

Define the project constitution: tech stack, conventions, constraints, Agent limits, and gap protocol. All globally-applicable decisions.

project-context.md
1

2. Design the Agent Chain

Define which Agents exist, what each one does, and how they pass work. Each with identity, responsibilities, limits, and gap protocol.

Agent Definitions
2

3. Generate Strategic Documents

PM Agent produces the PRD from the Solution Brief. Architect Agent produces the Architecture Document and promotes new Rules.

PRD + Architecture Document
3

4. Decompose into Skills

SM Agent breaks the solution into self-contained tasks with all the context an Agent needs to complete them.

Story Files
4

5. Validate through Outputs

Agents execute sample Skills under Rules. Stakeholders verify the output is a faithful expression of what was agreed.

Validated context system
5

The six artifacts​

This phase produces six artifacts that together form the operating system for Phase 4. Unlike previous phases, here artifacts are produced by both humans and AI Agents, and each has defined producers and consumers within the chain.

Context system quality​

A well-designed context system scores high across all five quality dimensions. Gaps in any dimension produce outputs that look correct but aren't faithful to what stakeholders agreed.

FidelityCompletenessSelf-containmentTraceabilityConsistency
Well-designed system
System with gaps
  • Fidelity — Outputs match the Problem Statement and Solution Brief. No invention, no reinterpretation.
  • Completeness — Every Rule, Agent, and Skill has all mandatory elements. No implicit context.
  • Self-containment — An Agent with a Skill + project-context.md can complete the task without asking anything.
  • Traceability — Problem Statement → Solution Brief → PRD → Rules + Skills. Every element traces back.
  • Consistency — Rules don't contradict each other. Skills inherit Rules faithfully. Agents respect their limits.

What this guide covers​

The operating system investment

If Context Phase consumes more time than the construction phase itself, you're likely over-specifying Rules or atomizing Skills excessively. If it consumes less than a third of construction time, there are implicit Rules and incomplete Skills. The optimal proportion is 30-50% of construction time. For a project that will take 4 weeks to build, invest 1-2 weeks in Context Phase. The payoff: construction becomes the phase of least friction.