Book a call
CORE COURSE18h 52m62 LESSONS

Driving the harness

The transferable half. Everything here is true of every harness — the switcher swaps the part that isn't.

  1. 01

    Foundations

    Where the model ends and the harness begins. Every other chapter assumes this one.

    2h 15m
    1. 01What a harness is, and why most "model quality" is harness qualityLESSON
    2. 02The agent loop: read, plan, act, observe — and where each step failsLESSON
    3. 03AGENTS.md, CLAUDE.md and the rest: what belongs in a memory fileLESSON
    4. 04Three gates, not one: permissions, approvals and sandboxingLESSON
    5. 05Reading a session like a profiler: where the context wentLESSON
    6. 06Verify it actually loaded: config that fails silentlyRECIPE
    7. 07Your first repo-shaped setup, start to finishBUILD
  2. 02

    Prompting an agent, not a chatbot

    The habits that transfer from chat are the ones that hurt most here.

    1h 22m
    1. 01Why agent prompting isn't chat promptingLESSON
    2. 02Specifying done: acceptance criteria the agent can check itself againstLESSON
    3. 03Give it the shape, not the stepsLESSON
    4. 04When to paste code and when to point at a pathLESSON
    5. 05Repair prompts: getting out of a bad turn without starting overRECIPE
    6. 06Asking for a plan you can actually rejectRECIPE
  3. 03

    Commands and skills

    One mechanism, two names — and an open standard several vendors now share.

    2h 29m
    1. 01Skills, commands and the merge: one mechanism, two namesLESSON
    2. 02Anatomy of a skill: frontmatter, arguments, and who invokes itLESSON
    3. 03Setting /goalsRECIPE
    4. 04Writing a skill that triggers when it should — and not when it shouldn'tRECIPE
    5. 05Writing a /review your team actually runsRECIPE
    6. 06Skills vs MCP: procedural knowledge or live tool accessLESSON
    7. 07Packaging so a new machine is one install awayBUILD
    8. 08Skill, rule, or memory file: picking the mechanismREFERENCE
  4. 04

    Tools and MCP

    Tools, not prompts. Wiring a model into the systems it has to touch.

    2h 46m
    1. 01MCP without the hype: what the protocol actually gives youLESSON
    2. 02Your first server: one tool, one schemaBUILD
    3. 03Tool descriptions are prompts — write them like itLESSON
    4. 04Scoping and auth: what a server should refuse to doLESSON
    5. 05Local or remote: where the trust boundary goesLESSON
    6. 06When not to reach for an MCPLESSON
    7. 07Too many tools: what a crowded tool list costs youLESSON
    8. 08Debugging a server the model keeps calling wrongRECIPE
    9. 09MCP in 2026: stateless, standardised, and what got deprecatedREFERENCE
  5. 05

    Subagents and fan-out

    Parallelism you can still review.

    2h 14m
    1. 01Subagents are a config object now, not a trickLESSON
    2. 02Subagent, background session, or agent team: three things people conflateREFERENCE
    3. 03Writing a subagent brief that returns data, not proseRECIPE
    4. 04Adversarial verification: a second agent whose job is to refute the firstRECIPE
    5. 05Judge panels: when three agents on one answer is worth itRECIPE
    6. 06Pipeline, barrier, loop-until-dry: orchestration shapesLESSON
    7. 07Worktree isolation for agents that write the same filesBUILD
  6. 06

    Hooks and guardrails

    Determinism around a non-deterministic core. At this speed, the gate is the design.

    2h 15m
    1. 01The moments in a session worth interceptingLESSON
    2. 02Blocking the commit the agent shouldn't makeRECIPE
    3. 03Format, lint and test as a hook, not as a plea in the promptRECIPE
    4. 04Failing loudly: surfacing a refusal the model will act onLESSON
    5. 05Permission fatigue and the middle pathRECIPE
    6. 06Secrets and the files an agent must never readLESSON
    7. 07A guardrail set for a team of sixBUILD
  7. 07

    Cost and control

    The loudest pain in the field, and the one the two chapters above create.

    1h 18m
    1. 01Where the money actually goesLESSON
    2. 02Reading your own usageRECIPE
    3. 03Budgets, caps and a kill-switch for unattended runsRECIPE
    4. 04What a twelve-agent sweep really billsREFERENCE
    5. 05Routing: when a cheaper model is the right answerLESSON
  8. 08

    Headless and CI

    The same harness with nobody at the keyboard.

    2h 9m
    1. 01Headless mode: what changes when there's no one to askLESSON
    2. 02Keyless CI: OIDC and workload identity before API keysLESSON
    3. 03An agent that triages issuesBUILD
    4. 04An agent that reviews pull requestsBUILD
    5. 05When it fails at 3amLESSON
    6. 06When not to automate itLESSON
  9. 09

    Doing this with other people

    Everything above, times a team.

    2h 4m
    1. 01Standardising a setup without a mandateLESSON
    2. 02Monorepos: scoping instructions to a subtreeLESSON
    3. 03The repo that explains itself to an agentBUILD
    4. 04Which agent for which job, and the handoff you still do by handLESSON
    5. 05Reviewing each other's agent configurationLESSON
    6. 06Measuring whether any of this is workingLESSON
    7. 07Anti-patterns we've watched teams adoptREFERENCE