Skip to content

Latest commit

 

History

History
35 lines (25 loc) · 4.8 KB

File metadata and controls

35 lines (25 loc) · 4.8 KB

Example-To-Mainline Source Bridge

中文 | English

First ask: which mainline step does this example sit on? Then ask: what real complexity does the source add at that step?

If you have not yet located the step, go back to the mainline source map. If you want the staged learning route, return to the Learning Paths.

Example → mainline step → next source hop

The source files listed below are first-hop reading targets, not exhaustive ownership boundaries for that behavior.

Example Mainline step What the example intentionally simplifies Which source files to open next
l1_startup.py Step 1 — Startup / mode routing compresses CLI parsing, global config, and runtime-mode dispatch into one startup skeleton main.tsx, setup.ts, cli/print.ts
l2_agent_loop.py Step 2 — Enter query / queryLoop compresses QueryEngine, turn state, and history assembly into one minimal loop QueryEngine.ts, query.ts
l3_tool_system.py Step 3 — Model output / tool selection compresses the real tool pool, schemas, and mode filtering into a small registry story Tool.ts, tools.ts, query.ts
l4_ui_ink.py Step 6 — State / UI update compresses the Ink component tree, state slices, and message rendering into one visible feedback loop bootstrap/state.ts, screens/REPL.tsx, components/Messages.tsx
l5_state_commands.py Step 6 — State / UI update compresses pending-message flow, command dispatch, and history sync into a simplified state machine screens/REPL.tsx, history.ts, commands/
l6_advanced.py Step 3 — Model output / tool selection compresses the real multi-agent / structured-output boundary into a “restricted tools + constrained worker” sketch tools/SyntheticOutputTool/, coordinator/coordinatorMode.ts, cli/print.ts
l7_permissions.py Step 4 — Tool execution compresses layered permissions, auto-approval, and dangerous-command checks into one gatekeeper tools/BashTool/bashPermissions.ts, tools/BashTool/bashSecurity.ts, query.ts
l8_streaming.py Step 3 — Model output / tool selection compresses provider event streams, chunk repair, and stop-reason handling into a readable async-generator view services/api/claude.ts, query.ts
l9_context_mgmt.py Step 7 — Next turn / exit conditions compresses prompt cache, context trimming, and memory extraction into one “how the next turn survives” story constants/prompts.ts, utils/context.ts, memdir/memdir.ts, services/extractMemories/extractMemories.ts

How to keep walking after the bridge

  1. Use the table to locate the example’s mainline step.
  2. Open the matching source-map section and verify the input / output pair.
  3. Open only the 1 to 3 source files listed in the table before branching out.
  4. After those files are stable in your head, decide whether to expand into the matching layer note or source-navigation.

Two common stitched routes

  • l1 -> l2 -> l8 -> l7 -> l4/l5 -> l9: read in the order a real request actually moves.
  • l3 -> l7 -> l6: lock onto tools and permissions first, then see how controlled multi-agent work reconnects to the mainline.