Section 06 · Anthropic Academy — Claude Code in Action · Builder

Claude Code in Action

Lifecycle events end-to-end, hook trust levels, subagents vs inline, CI surface, watermarks.

Learning objectives

  • Order all hook lifecycle events correctly in a single agent turn.
  • Pick hook vs CLAUDE.md by trust level (high trust = enforce; low trust = suggest).
  • Predict when a no-watermark loop will exhaust context and propose 80/95% gates.
  • Choose subagent vs inline for parallelism + isolation tradeoffs.
  • Reason about CI surface blast radius — autonomous agents need tighter permissions, not looser.