Tool-call parsing & the act step · Harness engineering

Forward-deployed / Learning zone

← All courses Harness engineering

← Track overview

Phase 00 — Setup & tooling

Phase 01 — LLM I/O foundations

Phase 02 — The agent loop

The agent loop from scratch Tool-call parsing & the act step Termination: stop conditions & max steps Turn history & conversation state Use it: the agent loop on the real SDK Error recovery inside the loop A streaming agent loop

Phase 03 — Tool engineering

Phase 04 — Context engineering

Phase 05 — Prompt & instruction architecture

Phase 06 — File & code operations

Phase 07 — Shell & sandbox execution

Phase 08 — Permissions & safety gating

Phase 09 — Memory & persistence

Phase 10 — Subagents & orchestration

Phase 11 — Planning & task management

Phase 12 — MCP & extensibility

Phase 13 — Retrieval & codebase understanding

Phase 14 — Reliability engineering

Phase 15 — Evals & testing the harness

Phase 16 — Observability & cost

Phase 17 — Security & alignment

Phase 18 — Production & deployment

Phase 19 — Capstone: build your own coding agent

Loading…

← PreviousThe agent loop from scratch PhaseOverview Next →Termination: stop conditions & max steps

Font

Size

100%

Margins