Skip to content

README.en.md

Latest commit

 

History

History
98 lines (62 loc) · 4.4 KB

README.en.md

File metadata and controls

98 lines (62 loc) · 4.4 KB

Claude Code Learning Paths

中文 | English

If docs/layers/ is the reference shelf, these path pages are the course index organized around the one-request lifecycle. The goal is not to finish every doc, but to make that request flow clear in stages.

First, remember what these path pages are for

These four stages are not a generic course list. They split the same request flow into four slices:

user input → startup / mode routing → queryLoop one-turn progression → model / tools / result flow-back → state / UI update → next turn or exit

Each time, you only need to answer one question: which slice of the request flow am I in, and what do I need to understand now?

What not to read yet

While following the path pages, do not:

  • treat docs/layers/ as a mandatory cover-to-cover course
  • dive into memory, runtime modes, multi-agent flows, or plugins first
  • switch into “full repository tour mode” just because you saw many file names
  • start linear reading of claudecode_src/src/ before P1 / P2 feel stable

When a page says “ignore this for now,” that is there to protect your sense of the request flow.

How the four stages map onto the lifecycle

Stage Request-flow position Only understand this in the current stage
P1 First Hour user input → startup / routing how the request actually enters the system
P2 Core Loop queryLoop one-turn core flow how model work, tool use, and result flow-back connect
P3 Source Reading direct reading of the core source path the key files and key symbols on the core path
P4 Advanced Architecture boundaries outside the core flow how permissions, memory, and runtime modes surround the core path

How the examples map onto the request flow and stages

Stage Request-flow slice Start with these examples What these examples show
P1 input → startup / routing l1_startup.py, l4_ui_ink.py how the request enters the system, and how the UI exposes early-state transitions
P2 queryLoop → model / tools / flow-back l2_agent_loop.py, l3_tool_system.py, l7_permissions.py, l8_streaming.py how one turn advances, how tools are called, how permissions intervene, and how streaming carries the full path
P3 direct reading of the core source path l5_state_commands.py, l9_context_mgmt.py how state and context support multi-turn execution and source tracing
P4 extension boundaries outside the core flow l6_advanced.py how coordinators, cost tracking, skills, and runtime boundaries hang off the core path

How to use these paths

  1. Start from P1 unless you are already comfortable with agent frameworks and Claude Code call chains.
  2. Finish one stage at a time instead of opening five deep-dive notes in parallel.
  3. In each stage, run the example first, then read the path page, then use the layer note and source-map to steady the request flow.
  4. If you get lost in the source, open the Source Navigation Guide immediately.
  5. At each stage, confirm why you are reading that page before you branch into extra material.

Recommended order

If you are new

Go in order:

P1 -> P2 -> P3 -> P4

If you already build agents

Start from:

  • P2 if you mainly want Claude Code’s loop / permission / streaming design
  • P3 if you are ready to read the core source directly

A standard study sequence

Try to use this sequence in every stage:

run example -> read path page -> read matching layer -> open source-map -> search symbols -> open source files

If you do not want to assemble that chain yourself, open:

When to stop and review

Each stage only asks you to answer the current page’s questions.

If you can already:

  • explain which slice of the request flow this stage covers
  • explain why that slice matters
  • locate the matching symbols
  • explain what engineering problem this layer solves

then continue.

If not, do not rush into the next page yet.

Read alongside