Phase 2 moves from on-demand invocation to scheduled and event-driven agent execution. This page documents the target architecture.

Status: Not yet implemented. This is a design document.

Goals

• Agents run on schedules without manual invocation
• Events (new PR, traffic spike, spend anomaly) trigger agent actions
• Reports are generated and delivered automatically

Scheduled Agents

Agent	Schedule	Output
CMO	Weekly	Traffic report (top pages, trends, recommendations)
CFO	Weekly	Spend report (cost by model, anomalies, optimization tips)
CTO	Daily	Blocked/stale issue summary
SEO	Monthly	Content gap analysis and internal linking audit

Event Triggers

Event	Agent	Action
New PR opened	CTO	Review scope, check for related Linear issues
Traffic spike on a post	CMO	Analyze referral source, recommend follow-up content
Spend spike	CFO	Alert with breakdown of what caused the increase
New blog post published	SEO	Check internal linking, suggest related posts

Orchestration Options

Claude Code cron tasks: Claude Code supports cron-style task scheduling. Simplest option for Phase 2 — define tasks that run agents on a schedule.

OpenClaw: Multi-agent orchestration framework. Could coordinate cross-agent workflows (e.g., CMO identifies topic, content team produces post). Adds complexity — only worth it if single-agent cron isn't sufficient.

GitHub Actions: For CI-adjacent triggers (PR opened, push to main). Can invoke Claude Code or API calls as workflow steps.

Prerequisites

• Phase 1 agents must be stable and producing useful output
• Need a delivery mechanism for reports (email, Slack, or PR comments)
• Cost guardrails to prevent runaway spend from automated invocations