Prompt ingestion
The REPL and input processor turn free-form user actions into structured runtime work.
src/screens/REPL.tsxsrc/utils/processUserInput/processUserInput.ts
Execution Flows
Trace the important chains
These flow maps help you understand Claude Code as a running system rather than a pile of files.
Diagram views
Interactive turn sequence
How one user turn moves through REPL, query, tools, and storage.
Source anchors
Query phase
The query loop coordinates model streaming, tool emission, and transcript updates.
src/query.tssrc/QueryEngine.ts
Persistence handoff
Transcript durability is a visible part of the normal turn lifecycle.
src/utils/sessionStorage.ts
Why this flow matters
Interactive Turn: input to streamed answer
The standard path for understanding how Claude Code works at runtime.
This flow connects REPL, input normalization, query orchestration, persistence, and rendering.
1
REPL captures prompt and queued commands
The UI gathers prompt text, attachments, and queued command metadata.
src/screens/REPL.tsxsrc/components/PromptInput/
2
Input is normalized and hook-aware
Plain text, slash commands, and attachments go through one normalization layer.
src/utils/processUserInput/processUserInput.ts
3
Query loop drives model and tool phases
The runtime streams model output, reacts to tool blocks, and manages retries or compaction.
src/query.tssrc/QueryEngine.ts
4
Messages are persisted before and after execution
Session storage keeps transcript durability and resume compatibility.
src/utils/sessionStorage.ts
5
REPL renders stream, tools, and notifications
The UI updates incrementally instead of waiting for the whole turn.
src/screens/REPL.tsxsrc/cli/print.ts