Agent Persona Team
Why this exists
John is a solo founder. He can't remember 14+ skill names, doesn't know which skill to invoke for which job, and shouldn't have to. This document names the agent collective as a team with roles, so scheduled routines and ad-hoc work have clear ownership.
John never invokes any of these roles directly. They run on schedules or are summoned by the Chief of Staff when needed. He interacts with the Morning Brief.
The roles (8 seats, mapped to existing skills)
| # | Role | Skills used | Who summons them | What they output |
|---|---|---|---|---|
| 1 | Chief of Staff | chief-of-staff + executive-assistant | Schedule + ad-hoc from any agent | Morning Brief; founder action items; escalation routing |
| 2 | Dev | feature-dev + raw implementation | PRs and bug reports | Feature branches, PRs, code changes (never on main/master) |
| 3 | QA | qa-orchestrator + ensure-solid + test-chatbot + test-voice + journey-runner | Schedule + every critical-path PR | Scorecards, Playwright artifacts, regression diffs |
| 4 | DBA | rag-architect + Supabase MCP + readiness-sot-check.ts | Schedule + migrations | Data integrity reports, knowledge-check output, migration PRs |
| 5 | Security | security-review skill | Every PR that touches auth/RBAC/Stripe/voice | PR review comments; moderation_violations weekly analysis |
| 6 | Doc Control | chief-of-staff (doc side) + derive scripts | Schedule (Mon 9am) | Freshness reports, stale-doc PRs, Docusaurus re-publishes |
| 7 | Marketing | marketing + church-outreach + seo-strategist + design | John asks for a campaign | Draft campaigns/copy (John reviews and sends); ad creative; SEO recommendations |
| 8 | CEO (John) | — | Self | Reads Morning Brief; approves drafts; does founder interviews; makes product/price decisions |
Daily cadence (all scheduled, zero John-invocation)
| Time (Toronto) | Role | Task | Output location |
|---|---|---|---|
| 06:00 | QA | /ensure-solid all | knowledge/readiness/reports/*.{md,json} |
| 06:15 | DBA | validate_product_knowledge() + pnpm derive --check + drift diff | Supabase query log + SOT report |
| 06:30 | QA | Synthetic funnel tests against prod URLs | Playwright HTML reports |
| 06:45 | Chief of Staff | Regression diff (new P0/P1 since yesterday) | ops_errors change-log |
| 06:50 | Chief of Staff | Email triage (both inboxes) | Drive receipts folder + draft queue + triage summary |
| 07:00 | Chief of Staff | Compose Morning Brief | morning_briefs table + email to john@churchwiseai.com + dashboard page |
Weekly cadence
| Day/Time | Role | Task | Output |
|---|---|---|---|
| Mon 09:00 | Doc Control | Knowledge freshness scan | FRESHNESS_REPORT.md updates + stale-doc PRs |
| Wed 09:00 | DBA | Customer health check (real customers only) | Churn-risk report → Brief |
| Fri 09:00 | Chief of Staff | Weekly retrospective draft | knowledge/retrospectives/<week>.md + link in Brief |
Monthly cadence
| Day/Time | Role | Task |
|---|---|---|
| 1st 05:00 | QA | Stripe 21-tier coupon validation (already scheduled) |
| 1st 07:00 | Chief of Staff | Monthly receipts roll-up in Brief (vendor spend breakdown + CSV for accountant) |
Escalation rules — when a role summons another
- QA → Dev: scorecard failure → Dev creates fix branch, opens PR, QA verifies
- DBA → Chief of Staff: knowledge drift that can't self-heal (derive fails) → Chief escalates to John in Brief
- Security → Chief of Staff: any unreviewed
moderation_violationswith severity ≥ 8 older than 24h → Brief escalation - Chief of Staff → John: any founder-only decision → Actions For You section of Brief
- Any role → Chief of Staff: cross-role coordination; Chief owns the Brief so it's the natural aggregator
What agents never do
- Commit directly to
main/master(always feature branches) - Push without
git pull --rebasefirst - Modify
critical_path: truecode files without attaching Playwright evidence (enforced by critical-path-gate workflow) - Scan the founder's personal email
- Make product/pricing/legal decisions (escalate to John)
- Redeploy the voice agent without founder present (voice agent is LOCKED)
Claude Code skill /schedule wiring (Week 1 deliverable)
Each of the 6 daily routines above is registered via the Claude Code /schedule skill on John's workstation. Configs live in C:/dev/.claude/scheduled-agents/. Configs are committed to the DEV repo so any agent can audit the schedule.
Phase 2 additions (post first-real-customer)
Add two new seats:
- Customer Success — new skill invoked weekly; reviews every real customer's usage + support history
- Growth — merges marketing + data-analyst responsibilities once MRR > $500/mo