Skip to content

l6-advanced.en.md

Latest commit

 

History

History
45 lines (30 loc) · 1.12 KB

l6-advanced.en.md

File metadata and controls

45 lines (30 loc) · 1.12 KB
Raw

Layer 6: Advanced Mechanisms

中文 | English

Core Question

How does Claude Code grow beyond the main loop into product-layer systems like skills, multi-agent work, memory, and cost tracking?

Run First

python examples/l6_advanced.py

Read First

  • skills/
  • coordinator/
  • tools/AgentTool/
  • tools/SyntheticOutputTool/
  • memdir/
  • cost-tracker.ts

Search Anchors

  • SYNTHETIC_OUTPUT_TOOL_NAME
  • createSyntheticOutputTool
  • MEMORY.md
  • cost

What To Notice

  • which capabilities are really “prompt reuse as a product feature”
  • how structured output locks the boundary around multi-agent work
  • why memory behaves like an index instead of a growing log dump

Demo vs Real Source

The demo compresses several advanced mechanisms into one file. The real source keeps them decoupled and lets the session orchestration compose them.

Questions To Answer

  1. What are the boundaries between skills, tools, and commands?
  2. Why should the coordinator not inspect worker intermediate state?
  3. Why is memory closer to a directory index than a long-term chat transcript?