Skip to main content

Exit Criteria: the Gate Review

Context Phase is not complete when the documents look polished. It's complete when the outputs of the system — Agents executing Skills under Rules — are recognized as correct by the stakeholders who defined the problem and the solution.

Gate 3 → 4: Passage conditions
  • ✅ The outputs of test runs (Agents executing Skills under Rules) are recognized as correct by stakeholders from Phases 1 and 2. They can look at the output and say: "Yes, this is a faithful expression of what we agreed."
  • ✅ The Rules pass the three-reads test: an Agent reading only the project-context.md can answer — What technologies do I use? What can't I do? What's the convention for any recurring pattern?
  • ✅ Every Agent has the five mandatory elements: identity, responsibilities, limits, Rules reference, and gap protocol.
  • ✅ Every Skill passes the self-containment test: an Agent with the Skill + project-context.md can complete the task without asking anything.
  • ✅ The PRD has total traceability to Phase 1 and 2 artifacts. Every section traces to the Problem Statement or Solution Brief.
  • ✅ The Architecture Document includes ADRs and has generated global Rules in the project-context.md.
  • ✅ The QA Agent has validated every document in the chain: fidelity to inputs, completeness, consistency with Rules.
  • ❌ Do not advance if test outputs are not recognized as faithful to the Problem Statement and the Solution Brief.
  • ❌ Do not advance if there are Skills that depend on implicit context not present in Rules or in the Skill itself.
  • ❌ Do not advance if there are implicit Rules not written in the project-context.md — if the team "knows" something but it isn't documented, it doesn't exist for the Agents.

These Exit Criteria exist because Context Phase is the last checkpoint before construction. Every gap that passes this gate becomes invisible Context Debt during Phase 4: Agents make confident decisions based on incomplete information, producing outputs that look correct but aren't faithful to what stakeholders agreed.

The validation question​

The fundamental question at the Gate Review is not "Are the documents complete?" — it's "Are the outputs correct?"

This distinction matters. A project-context.md can look thorough, Agent Definitions can seem well-structured, and Skills can appear detailed. But the only real test is running the system and evaluating what it produces. Documents are the means; outputs are the evidence.

The validation works like this:

  1. Select 3-5 representative Skills that cover different areas of the project.
  2. Have the corresponding Agents execute them under the current Rules.
  3. Present the outputs to stakeholders — the same people who validated the Problem Statement and the Solution Brief.
  4. Ask: "Is this recognizable as a faithful expression of what we agreed?"
  5. If yes for all samples: the context system is validated.
  6. If no for any sample: diagnose where the failure is (Rules, Agent, Skill, or interaction) and iterate.

After the gate: transition to Phase 4​

When the team completes the Gate Review, the project transitions from design to execution. But unlike traditional projects, the construction phase doesn't start with a briefing — it starts with a complete operating system.

Every development Agent already knows:

  • What technologies to use (Rules, Section 2)
  • What conventions to follow (Rules, Section 3)
  • What's prohibited (Rules, Section 4)
  • What decisions have been made and why (Rules, Section 8)
  • What to do when information is missing (Rules, Section 7)

And every Skill they'll execute contains:

  • What to achieve (Objective)
  • What they need to know (Specific context)
  • How to know they're done (Acceptance criteria)
  • What must be complete first (Dependencies)
  • What additional limits apply (Constraints)

This is why, with a well-designed context system, construction tends to be the phase of least cost and least friction. The hard work was done in Phases 1-3. Phase 4 is execution.

The context doesn't freeze

Passing the Gate Review doesn't mean the context is frozen. During Phase 4, the context continues evolving:

  • Construction reveals ambiguities → Skills are refined.
  • A local decision during a Skill should be global → new Rule.
  • An Agent needs unforeseen capabilities → Agent definition is updated.
  • An active assumption is refuted → affected Rules and Skills are adjusted.

Every change is documented in the Decision Log. The system is alive, but every mutation is deliberate and recorded.