Learn Claude Code

Learning Path

s01 to s12: Progressive Agent Design

Layer Legend

Tools & Execution
Planning & Coordination
Memory Management
Concurrency
Collaboration
Observability & Evaluation
Safety & Governance
01
s01Single-tool agent loop

The Agent LoopBash is All You Need

98 lines of code1 tools

The minimal agent kernel is a while loop + one tool

Learn More
02
s02Tool dispatch map

ToolsOne Handler Per Tool

154 lines of code4 tools

The loop stays the same; new tools register into the dispatch map

Learn More
03
s03TodoManager + nag reminder

TodoWritePlan Before You Act

212 lines of code5 tools

An agent without a plan drifts; list the steps first, then execute

Learn More
04
s04Subagent spawn with isolated messages[]

SubagentsClean Context Per Subtask

206 lines of code5 tools

Subagents use independent messages[], keeping the main conversation clean

Learn More
05
s05SkillLoader + two-layer injection

SkillsLoad on Demand

234 lines of code5 tools

Inject knowledge via tool_result when needed, not upfront in the system prompt

Learn More
06
s06micro-compact + auto-compact + archival

CompactThree-Layer Compression

290 lines of code5 tools

Context will fill up; three-layer compression strategy enables infinite sessions

Learn More
07
s07TaskManager with file-based state + dependency graph

TasksTask Graph + Dependencies

251 lines of code8 tools

A file-based task graph with ordering, parallelism, and dependencies -- the coordination backbone for multi-agent work

Learn More
08
s08BackgroundManager + notification queue

Background TasksBackground Threads + Notifications

230 lines of code6 tools

Run slow operations in the background; the agent keeps thinking ahead

Learn More
09
s09TeammateManager + file-based mailbox

Agent TeamsTeammates + Mailboxes

386 lines of code10 tools

When one agent can't finish, delegate to persistent teammates via async mailboxes

Learn More
10
s10request_id correlation for two protocols

Team ProtocolsShared Communication Rules

466 lines of code12 tools

One request-response pattern drives all team negotiation

Learn More
11
s11Task board polling + timeout-based self-governance

Autonomous AgentsScan Board, Claim Tasks

543 lines of code14 tools

Teammates scan the board and claim tasks themselves; no need for the lead to assign each one

Learn More
12
s12Composable worktree lifecycle + event stream over a shared task board

Worktree + Task IsolationIsolate by Directory

737 lines of code16 tools

Each works in its own directory; tasks manage goals, worktrees manage directories, bound by ID

Learn More
13
s13run_tests + evaluate tools with structured results

Agent EvalsTDAD + LLM-as-Judge

256 lines of code5 tools

An agent that can't verify its own work is just guessing; verifiable output is the key to success

Learn More
14
s14Chaining, Routing, Parallelization, Orchestrator-Worker, Evaluator-Optimizer

Workflow PatternsFive Orchestration Strategies

335 lines of code4 tools

Don't build agents — build workflow patterns; start simple, add complexity only when needed

Learn More
15
s15ContextBudgetManager + LayeredConfig + AgenticMemory

Context EngineeringBudget, Not Compress

291 lines of code6 tools

Context Engineering > Prompt Engineering; proactive budget allocation, not reactive compression

Learn More
16
s16Planner/Generator/Evaluator with Fresh Context Window

Long-Running HarnessGAN-Style Three Agents

267 lines of code5 tools

Separate the generator from the evaluator; fresh context per iteration solves context anxiety

Learn More
17
s17MCPServer/MCPClient with runtime tool discovery

MCPStandard Tool Protocol

230 lines of code7 tools

MCP is USB for AI tools — standardized discovery and invocation across any service

Learn More
18
s18PermissionManager (SAFE/ASK/DENY) + Sandbox

Auto ModePermission Tiers + Sandbox

234 lines of code6 tools

Safety should not be the enemy of efficiency; 95% auto-approve, 5% human oversight

Learn More
19
s19think tool — returns input as-is

Think ToolA Tool That Does Nothing

163 lines of code4 tools

The simplest tool can be the most useful; giving the model permission to pause improves quality

Learn More
20
s20TaskBoard + parallel Workers + atomic claiming

Parallel TeamsFile-Lock Task Board

264 lines of code6 tools

The key to scaling is decoupling; task board + file locks = linear scalability

Learn More
21
s21Structured output + diff-based editing + poka-yoke

Tool DesignACI Principles

280 lines of code6 tools

Good tools are hard to misuse; tool design impacts agent quality more than prompt design

Learn More

LOC Growth

s01
98
s02
154
s03
212
s04
206
s05
234
s06
290
s07
251
s08
230
s09
386
s10
466
s11
543
s12
737
s13
256
s14
335
s15
291
s16
267
s17
230
s18
234
s19
163
s20
264
s21
280