Development Process

Evidence-based agent workflow

The new process relies on two sources of requirements only: recorded meetings with screen sharing and emails. Everything else is auxiliary communication, not a source of work. AI agents prepare proposals, specs, and tests from this evidence, but every specification must pass a manual review step (architect/PM) before development starts.

graph TD
    Evidence["Meeting recording<br/>+ emails"] --> Intake["Evidence intake<br/>Transcript + quotes"]
    Intake --> Architect["Architect Agent<br/>E2E design"]
    Architect --> Spec["OpenSpec draft<br/>manual review"]
    Spec --> Sprint["Sprint backlog<br/>user stories"]
    Sprint --> Build["Dev/QA Agents<br/>PHPStan+PHPUnit+Lint"]
    Build --> Click["UI click tests<br/>click-test repo"]
    Click --> Deploy["Deploy + spec archive"]
    Deploy --> Change["Iterative changes<br/>Change Proposal"]
    Change --> Intake

Process phases

Phase	Description
Evidence Intake	Collecting and validating evidence
Architect & Specification	Design and review
Sprint Build	Implementation with quality gates
Testing	Unit/integration + UI click tests
Release	Deployment, archiving, and iteration

Evidence intake (mandatory rules)

• Allowed sources: (1) meeting recording with screen share, (2) email thread. Nothing else.
• Transcript: every meeting has a transcript with timestamps and a link to the video; every email has the message-id stored.
• Traceability: every requirement must include source: meeting:<date>#<timestamp> or source: email:<message-id>.
• Validation: no work starts without attached evidence. If evidence is missing, the request goes back to intake.

---

➡️ Continue to Evidence Intake.