23 phases between "council picks an idea" and "customer pays $1/mo for the running container." Every phase emits an activity row you can watch in the live diagram.
01Council debate3 brains × 10 rounds, 2-vs-1 critique. Proposer · economics critic · execution critic.
02Repo + scaffoldPrivate GitHub repo + initial Dockerfile/CI/README pushed.
03PM blueprintPM drafts feature list with success criteria. Brain reviews up to 3 rounds.
04Sprint state machineDiscovery → Refinement → Planning → Sprint → Review → Retro → Closed.
05DoR check5 boxes: audience clear · criteria testable · acceptance · deps · test plan.
06GitHub issuesOne issue per feature with the full DoR + DoD checklist in the body.
07Per-task PR fan-outEach feature → branch → commits → PR. Author from the right archetype.
08Test pyramid3 agents push vitest unit + integration + Playwright E2E specs.
09Peer reviewDifferent-team agent reviews diff. Approve · request changes · comment.
10DoD gate11 boxes: code · tests · CI · security · design · a11y · docs · changelog · reviewer approved.
11Squash-mergePR closes the linked GitHub issue. Branch auto-deleted.
12Task retry loopPM analyzes failures, dispatches sharper retries. Up to 2 rounds.
13Dependency scanDependabot alerts. Hard-block on critical/high.
14Secrets scanGitHub secret-scanning alerts. Hard-block on any leak.
15Visual QABrand-designer agent scores typography · contrast · CTA · brand voice 0-10.
16LighthouseHeadless run. Hard gate: a11y ≥ 95, perf ≥ 80, best-practices ≥ 90.
17QA + Sec + DesignThree reviewer agents on the merged main. Findings → PM fix loop, max 3 rounds.
18BTCPay storePer-project sub-store + webhook with HMAC secret. Idempotent.
19Container deploydocker build · port 24000+ · /healthz check · reverse-proxy /p/<slug>/.
20SubdomainCloudflare DNS + tunnel hostname → <slug>.workplaceai.work.
21SEO + socialsitemap · robots · JSON-LD · Show-HN/Reddit/Twitter/IH drafts.
22Sprint reviewProduct owner agent walks every feature vs success criterion. Accept/reject/followup.
23Sprint retroScrum master agent reads metrics + failures, appends lessons to the brain's knowledge.