Lesson 25 / 25
Revision: Cheat Sheet and Self-Check
Review the patterns, tools and safety habits from the whole course.
Cheat sheet
Patterns: chain, route, parallelise, orchestrate, evaluate-and-refine; start simple. Skills: SKILL.md with a sharp description, scripts for exact work. Subagents: isolate noisy work, give minimal tools. Context: keep it small, use CLAUDE.md and progress files. Hooks: enforce rules as code. Safety: least privilege, distrust inputs, step and cost limits, human checkpoints. Quality: verification commands and evals.
Questions interviewers ask
Be ready to explain: workflow versus agent and when each fits, why a hook is stronger than a prompt rule, what progressive disclosure means for skills, how a subagent protects context, and how you would stop a runaway loop.
Quick check: A rule must never be broken, such as "do not push to main". What enforces it best?
- A polite line in the prompt
- A hook or permission rule that blocks it
- Hoping the model remembers
- A longer CLAUDE.md
Answer
A hook or permission rule that blocks it — Enforcement in code works even when the model forgets or is tricked.
Quick check: Which statement about subagents is correct?
- They share the main conversation's full memory
- They work in their own context and return a summary
- They cannot use tools
- They always make tasks cheaper
Answer
They work in their own context and return a summary — Isolation is their purpose; they still cost tokens, so use them where it pays off.
Quick check: When should you add an agent loop instead of a fixed workflow?
- Always, as the first step
- When steps cannot be predicted and simpler designs fall short
- When you want fewer tests
- When latency must be zero
Answer
When steps cannot be predicted and simpler designs fall short — Agents add cost and unpredictability, so use them only where flexibility is needed.