Course Map

Agentic Harness Engineering (in less than 1 hour)

Why the harness - not the model - decides your results, and how to engineer one. A course for developers new to coding agents.

Sprint Zero builds the shared vocabulary (knowledge). Chapter 1 turns it into a practice (skills). Chapter 2 goes deep on spec-driven development; Chapter 3 scales the harness to real, safe, autonomous work; Chapter 4 measures and evolves it; Chapter 5 takes it to many agents and a whole team; Chapter 6 is the capstone - assemble it, run it end to end, and make it yours. Read in order; each lesson is a few minutes. Keep the Glossary open in a tab.

Chapter 0 · Sprint Zero

Common ground: what the pieces are and why they matter.

  1. 0.1The Model Landscape Frontier vs open-weight, the narrowing gap - and why model sameness makes the harness the differentiator.
  2. 0.2What a Harness Is (and How It Evolved) Agent = Model + Harness, the reason-act-observe loop, and why harnesses shrink as models grow.
  3. 0.3Spec-Driven Development & the Toolkit Trio Spec as source of truth; Spec Kit vs OpenSpec vs agent-skills - three ways to bring SDD to your agent.

Chapter 1 · The Ratchet & the Practice Loop

The core skill and the habits that compound it.

  1. 1.1The Ratchet Fix the harness, not the prompt: every mistake becomes a permanent fix.
  2. 1.2CLAUDE.md as a Pilot's Checklist Keep the rules file lean and earned - a checklist, not a style guide.
  3. 1.3Hooks, Not Instructions When a rule must hold every time, enforce it with a script.
  4. 1.4Spec Before Code Pin the decisions before the agent writes - states, not activities.
  5. 1.5Context Firewalls & Subagents Keep the window clean; use subagents to isolate the messy work.
  6. 1.6Cross-Agent Review A second, independent agent catches what the first one missed.

Chapter 2 · Spec-Driven Development in Depth

A deep dive on the spec frameworks - with Spec Kit as the main event.

  1. 2.1The SDD Mindset & the Trio Why spec-driven development pays, and the three frameworks that deliver it.
  2. 2.2Spec Kit, End to End Walk GitHub Spec Kit's full loop, from installing the CLI to generated code.
  3. 2.3Spec Kit in Practice The analyze quality gate, customization, and knowing when it is overhead.
  4. 2.4OpenSpec & agent-skills The two lighter alternatives - a lightweight framework and a drop-in skill.
  5. 2.5Choosing & Adopting a Spec Framework A decision aid for picking Spec Kit, OpenSpec, or agent-skills.

Chapter 3 · Scaling & Trusting the Harness

Running agents for real work - sharp tools, safe tools, solid ground, eyes on cost.

  1. 3.1Tools: Fewer and Sharper A small set of focused tools beats a big overlapping pile.
  2. 3.2MCP & Tool Safety The tools you plug in are trusted text the model obeys - vet them.
  3. 3.3Filesystem, Git & Sandboxes Durable memory, an undo button, and a safe place to run code.
  4. 3.4Long-Horizon Autonomy Keep a long, many-step run on track: plan files, done-conditions, resets.
  5. 3.5Cost, Observability & HaaS Know what it costs, watch what it does, know when to stop tuning.

Chapter 4 · Measuring & Evolving the Harness

You can't improve what you don't measure - build a yardstick, read failures, close the loop.

  1. 4.1Why Vibes Aren't Enough "It feels better" is not evidence; to improve a harness you must measure it.
  2. 4.2Building an Eval Set A handful of your own real tasks with known-good outcomes becomes your yardstick.
  3. 4.3Reading Failure, Not Pass Rates A pass rate tells you IF; the failures tell you WHY - where the next fix lives.
  4. 4.4The Self-Improving Loop The ratchet, automated: the agent edits its own harness from what it learns.
  5. 4.5Knowing a Change Helped A/B against your eval set; keep the change only if it clears the noise.

Chapter 5 · Multi-Agent & Team Harnesses

Scaling across agents and across people - orchestrate deliberately, share the harness, fight drift.

  1. 5.1When to Add a Second Agent More agents isn't better; add one only for isolation, parallelism, or diversity.
  2. 5.2Orchestration Patterns Fan-out, pipeline, judge panel, loop-until-done - a shape for every case.
  3. 5.3Worktrees & Parallel Isolation Give each parallel agent its own copy so they can't clobber each other.
  4. 5.4The Team Harness Commit rules, hooks, and skills so the whole team inherits one earned harness.
  5. 5.5Fighting Drift & Governance A shared harness rots; keep it lean by reviewing rules like code.

Chapter 6 · Capstone: Build Your Harness

Pull the whole course into a real, adopted practice - assemble it, run it end to end, make it yours.

  1. 6.1Your Starter Harness The minimum viable harness you can stand up today, one piece per chapter.
  2. 6.2One Feature, End to End Watch a single feature travel the whole loop so every lesson clicks together.
  3. 6.3The First-Week Ratchet Plan A five-day plan that turns the discipline into a daily habit.
  4. 6.4Adapt, Don't Copy Keep the universal principles; tune the rest to your stack, team, and model.
  5. 6.5Staying Current & Your Capstone The ratchet mindset outlives every model - plus a one-month capstone challenge.
I'm your teacher. Start at 0.1 and ask me anything as you go. Dog-fooding this to teach others? Tell me which lesson ran too dense and I'll simplify it.
Glossary