Skip to slide
160 / 170

CH 17 · Building the Full Mini-Harness · 01/11

Read as one page

Building the Full Mini-Harness

A working mini coding-agent harness that assembles the loop, layered prompts, approval-gated tools, and memory into one terminal agent.

11 slides · 4 min read

Contour map for Building the Full Mini-Harness

In this chapter

  1. 01Architecture overview
  2. 02File structure
  3. 03Step 1: the model seam
  4. 04Step 2: wire the loop
  5. 05Step 3: the entry point
  6. 06Step 4: run it
  7. 07How the chapters show up at runtime
  8. 08Where to take it next
  9. 09Connecting to the bigger picture
  10. 10Key takeaways

Time to assemble. Every MVP from Chapters 2 through 14 was a real part of one system, and this chapter snaps them together into mininharness, a small but genuine terminal coding agent. It runs the loop, builds layered prompts, manages context and budget, runs tools behind an approval gate, loads project memory and skills, and can delegate to a subagent. It is a few hundred lines, small enough to read in a sitting, and shaped like the production systems this guide dissects. The only thing it does not ship with is a real model; you drop in an API call (Anthropic, OpenAI, or a local model) in one place, and everything else is the harness, which, as Chapter 1 promised, is 95% of the work.

Finished this idea?

Your learning progress stays in this browser.