Learning Objectives & Key Takeaways
0 / 4 Completed- Understand Claude Code core architecture and runtime execution model
- Audit real-world latency, throughput benchmarks, and memory footprint
- Implement zero-downtime deployment patterns with resilience guards
- Verify code snippets and architect blueprints against production SLOs
- CLAUDE.md shapes behavior but never enforces a boundary; only permissions and hooks do that.
- PreToolUse hooks run before every permission prompt except EndConversation and can deny calls outright.
- In promptless or background sessions, an undecided hook call is denied by default, not allowed.
- Deny rules beat allow rules every time, even under bypassPermissions mode.
- Scope MCP access narrowly per subagent instead of granting the mcp__* wildcard to everyone.
The 2 AM Incident: A Background Agent and a Missing Deny Rule
A platform team had a nightly subagent cleaning stale build caches across a monorepo, running non-interactively from a cron job with the Bash tool broadly allowed and nothing in the deny list.
One night a shell variable resolved empty before expansion, turning a scoped rm -rf $BUILD_DIR/* into something far wider, and because the session was running unattended there was no permission prompt for a human to catch on the way past.
- The Configuration Stack, In the Order Claude Code Expects
- CLAUDE.md and Slash Commands: Guidance the Model Reads, Not a Fence
- Hooks: The Only Thing That Still Runs When Nobody Is Watching
- A Locked-Down Subagent, Its Hook, and Its Narrow MCP Access
- MCP Servers and Permission Order: Deny, Then Ask, Then Allow
Checking your membership…
Members read every Tech Buzz story in full, with code examples, audio and the discussion, plus every course, quiz and interview set in the Learning Dashboard.



