For developers using Claude Code who want to move past trial-and-error. This guide is built from the patterns that actually work when you're: - Running multi-agent workflows in parallel - Managing your context window like a scarce resource - Making fast calls on "should this be inline or dispatched?" - Routing tasks to the right model tier (cheap vs. frontier) - Handling subagent failures gracefully It's not tutorial-level. Not "10 tips." It's how someone operates at the edge case of agentic Claude Code — what to do, when to do it, and why. Contents: - Dispatch decision trees (which agent type, when) - Model routing for cost + reasoning (when a cheap model saves tokens, when a frontier model pays for itself) — including a hard sensitivity gate for what never leaves your control - Context economy discipline (11 concrete practices) - 6 reusable prompt templates (research, architecture, code review, data transformation, debugging, escalation) plus a ready-to-paste CLAUDE.md starter-rules set - Output formatting that scales (tables, progress bars, structured thinking) - Risk gates: what needs human confirmation, what you execute autonomously - Real workflow examples with failure/recovery texture (sanitized, generalizable) Full table of contents: 1. Principles: Agentic Thinking 2. Dispatch Triage: Agent Type Decision Tree 3. Model Routing: Cost + Capability Trade-offs 4. Context Economy: Managing Your Window 5. Reusable Prompt Templates 6. Output Formatting That Scales 7. Risk Gates & Confirmation 8. Real Workflow Examples One-time $79 purchase. Useful for: senior developers, agentic AI builders, solo operators running multi-tool orchestration.