Prompt file imported from koldovsky/2026-fwdays-agentic-greenfield-task-live (
.codex/prompts/project-factory-init.md). Copyright stays with the author.
Execute the Project Factory init procedure. Canonical steps: .project-factory/commands/init.md
(summarized in AGENTS.md, which Codex reads natively as the rules).
Install the per-project loop — the deterministic scripts/check-*, git hooks,
CI, OpenSpec, and filled templates — idempotently and non-destructively
(Gate G0). Then follow .project-factory/MASTER-PROMPT.md from Phase 1 under
.project-factory/checklists/quality-gates.md.
Codex subagents have no parallel Workflow fan-out, so run the review/eval/spec
passes sequentially with fresh context (maker ≠ checker) — but still produce
the real evidence artifacts; the gates check artifacts, not process. For Phase 6
recordings use LOCAL headless Playwright (npm i -D @playwright/test && npx playwright install chromium && node scripts/record-demos.mjs), never an in-app
browser connector, then check-recordings + check-a11y + vision-verify. The
scripts, gates, specs, and evidence are identical to the Claude Code path. See
.project-factory/docs/portability.md.