Interview index: 232 source-backed questions
An index of 220 review questions from 22 architecture chapters, plus 12 checks from the two engineering-method chapters. Every entry leads back to a worked answer and source trail.
Choose a reading path
Section titled “Choose a reading path”Four ways to use this page:
- By chapter order: read chapters 1 through 22. Each has 10 questions, and later chapters go deeper.
- By topic cross-cut: use the topic quick-find table to see the same question class from different angles.
- As an interview question bank: every chapter link lands on its appendix, where the complete questions, answers, and follow-ups live.
- As a self-checklist: when building your own agent, find the chapter matching the design choice ahead, then audit the decision against those 10 questions.
Agent fundamentals
Section titled “Agent fundamentals”Themes: harness definition, agent loop, context, tool, verifier. These 50 cover the vocabulary and trade-offs of agent fundamentals.
| Chapter | Topic | Link |
|---|---|---|
| §1 | Agent Harness definition, concept boundaries, picking among the four systems | 01 · Overview |
| §2 | Agent loop five phases (perceive / plan / act / observe / verify), turn data structure, prompt assembly, context compaction, error recovery, tool dispatch, verifier | 02 · Agent Loop |
| §3 | Context systems: full spectrum from markdown to 10-layer assembly; how prefix cache holds up | 03 · Context system |
| §4 | Tool system: protocol, registration, parallelism, security gating | 04 · Tool system |
| §5 | Verifier: hard / soft / give-up tiers + stop detection | 05 · Verifier |
Coding-agent engineering
Section titled “Coding-agent engineering”Themes: real coding-agent engineering details. Patch algorithm, shell invocation, git workflow, code review automation. These tell you what extra trouble a coding agent has over a chat agent.
| Chapter | Topic | Link |
|---|---|---|
| §6 | File edit + patch algorithm: apply_patch V4A multi-version diff merge | 06 · File edit patch |
| §7 | Shell execution: execpolicy / sandbox / approval mode | 07 · Shell execution |
| §8 | Git workflow: worktree / branch / commit automation | 08 · Git workflow |
| §9 | Code review: /security-review + LLM-as-reviewer pattern | 09 · Code review |
Concurrency and runtime
Section titled “Concurrency and runtime”Themes: subagent, session lifecycle, permissions, sandbox, multi-channel entry. These 50 are the engineering foundation for keeping the agent stable in production.
| Chapter | Topic | Link |
|---|---|---|
| §10 | Subagent: fork / inline boundary, context isolation | 10 · Subagent |
| §11 | Session lifecycle: start / resume / interrupt / persist | 11 · Session lifecycle |
| §12 | Permissions and approvals: AskForApproval 4 levels + autoMode rules + INSTALL_POLICY 12-cell matrix | 12 · Permissions |
| §13 | Sandbox execution: macOS seatbelt / Linux bwrap+seccomp+landlock / Windows | 13 · Sandbox execution |
| §14 | Multi-channel entry: CLI / TUI / IDE / web / cron | 14 · Multi-channel entry |
Evolution, safety, and long-running systems
Section titled “Evolution, safety, and long-running systems”Themes: observability, memory, skills, scheduled tasks, self-improvement, security, todo lists, execution-state surfaces. These 80 are the core for long-running agents that stay recoverable and safe.
| Chapter | Topic | Link |
|---|---|---|
| §15 | Observability and cost: token / latency / tool.loop detection | 15 · Observability and cost |
| §16 | Memory system: 4 MemoryTypes, frozen snapshot, decay, prefix cache | 16 · Memory |
| §17 | Skill system: 8 crates / skillify / skill-scanner / 12-cell trust matrix | 17 · Skills |
| §18 | Cron and background tasks: tick + scheduler lock + isolated-agent | 18 · Cron background tasks |
| §19 | Self-improvement: Phase1+2 / skillify / passive indexing / in-turn explicit writes | 19 · Self-improvement |
| §20 | Security: sandbox / external-content / redact / cosign provenance | 20 · Security |
| §21 | Todo List: update_plan / TodoWrite / runtime progress / compaction recovery | 21 · Todo List |
| §22 | Execution State Surfaces: Plan / Todo / tool progress / task progress / terminal status / away summary routing | 22 · Execution State Surfaces |
Two engineering-method chapters
Section titled “Two engineering-method chapters”Topics: Loop Engineering and Graph Engineering. These are portable methods rather than four-system comparisons: how budgets are set, what edges carry, and where guardrails live. The material is grounded in mini-swe-agent, smolagents, crewAI, and open-source Codex.
| Chapter | Topics | Link |
|---|---|---|
| §23 | Loop engineering: five components, 3-dimension budgets, grace call, error taxonomy, breaker, trajectory, verifier hardness | 23 · Loop Engineering |
| §24 | Graph engineering: four node types, five edge elements, permissions-only-narrow, split decision, freezing, runaway protection | 24 · Graph Engineering |
Browse by question type
Section titled “Browse by question type”This section re-slices the 220 questions from the architecture chapters; the method chapters’ 12 checks remain in the section above. The same engineering problem appears from different angles in different chapters, so horizontal reading is often faster.
Concepts and boundaries
Section titled “Concepts and boundaries”Aligning vocabulary across the four systems. Common in “what’s the difference between A and B” or “why this split”.
| Source | Question |
|---|---|
| §1 Q1 | Agent Harness vs Agent model |
| §2 Q1 | Is “turn” a physical or logical unit |
| §4 Q1 | Tool / Function / Skill boundary |
| §10 Q1 | Subagent / fork / inline distinction |
| §17 Q1 | Skill / prompt template / tool / agent boundary |
| §18 Q1 | Cron job / background task / long-running task differences |
| §21 Q1 | Todo list vs Plan Mode boundary |
| §22 Q1 | Execution-state surface vs todo-list boundary |
Trade-off judgments
Section titled “Trade-off judgments”Why not B, with concrete context. Common in “why did Codex use X while Hermes used Y.”
| Source | Question |
|---|---|
| §2 Q4 / §3 Q6 | When to use prefix cache vs recompute |
| §11 Q3 | Whether session persistence needs versioning |
| §13 Q5 | Why sandbox doesn’t ban all network |
| §15 Q4 | How observability data avoids exploding disk |
| §16 Q5 | Memory limit in tokens vs chars |
| §18 Q3 | Cron failure auto-disable vs retry |
| §19 Q3 | OpenClaw halfLifeDays=30 derivation |
| §20 Q3 | fail_open vs fail_closed |
| §21 Q6 | Lightweight TodoWrite vs file-backed Tasks |
| §22 Q6 | Why progress display needs per-platform settings |
Implementation details
Section titled “Implementation details”Concrete APIs, data structures, source pointers. Common in “why does this code do that” / “how does it handle X at runtime.”
| Source | Question |
|---|---|
| §2 Q7 / §6 Q4 | apply_patch V4A multi-version diff merge |
| §3 Q8 | Hermes 10-layer prompt assembly ordering |
| §4 Q6 | dispatchToolUseBlocks parallel dispatch implementation |
| §5 Q5 | Codex goals.rs convergence judgment |
| §12 Q4 | AskForApproval 4 levels mapping to UI |
| §16 Q8 | Hermes _file_lock fcntl/msvcrt cross-platform |
| §17 Q6 | Hermes 12-cell INSTALL_POLICY calculation |
| §20 Q1 | OpenClaw random 8-byte ID wrapping |
| §21 Q8 | Todo replace vs merge API trade-off |
| §22 Q3 | MCP progress vs tool result protocol boundary |
Security questions
Section titled “Security questions”Injection / poisoning / secrets / supply chain. Common in “how does this attack get defended” / “why design it this way.”
| Source | Question |
|---|---|
| §12 Q9 | Keeping user password out of prompt |
| §13 Q7 | Common sandbox-escape paths |
| §17 Q7 | OpenClaw skill-scanner 3 severities |
| §19 Q5 | Hermes reasoning for blocking invisible unicode |
| §19 Q10 | 6-layer security defense stack |
| §20 Q1 | external-content random ID forgery prevention |
| §20 Q2 | tirith exit code as source of truth |
| §20 Q5 | Codex memory prompt explicit “treat as data” |
| §20 Q10 | 5-layer general security defense |
| §21 Q9 | Why todo completion needs verification nudges |
| §22 Q8 | Why progress surfaces still need no-progress detectors |
Engineering practice
Section titled “Engineering practice”How to actually build your agent. Common in “what’s the minimum production needs” / “what’s the minimum viable implementation.”
| Source | Question |
|---|---|
| §1 Q9 | Scenarios where heavy harness is wrong |
| §5 Q10 | 5 iron rules for verifier design |
| §10 Q10 | General subagent design checklist |
| §15 Q10 | General observability framework |
| §16 Q10 | General 5-layer memory architecture |
| §17 Q10 | General 6-layer skill system |
| §18 Q10 | General 7-layer cron system |
| §19 Q10 | General 6-layer self-improvement defense |
| §20 Q10 | General 5-layer security defense |
| §21 Q10 | Minimum 5-layer todo/progress surface |
| §22 Q10 | Minimum execution-state router fields |
After the questions
Section titled “After the questions”After 220 questions, the next step is turning insight into reusable engineering capability. Two paths:
- Build directly: pick a chapter’s implementation recipe and follow it to a minimum viable version.
- Crystallize into a skill: use the companion
~/.claude/skills/build-your-own-agent/skill to invoke the best practices distilled from these 220 questions on demand. See the<BuildRecipe />block at the end of each chapter.