Module 01: Install Nool and onboard a repository
Install the CLI, build the ledger from existing Git history with nool init --from-git, map the architecture with nool discover features, and pass nool doctor. Nothing in .git changes; the ledger lives beside it and mirrors back to Git.
Module 02: Your first governed change
Run the loop by hand: nool context, nool announce intent with a leased scope, nool propose under that intent, nool solidify, then read it back with nool log and nool why. Do it by hand once; from then on the agent runs the same four commands.
Module 03: Memory that survives the session
Record scoped findings with nool learn, retrieve them from a fresh session with nool findings and nool context, supersede a rule when it changes, and hand a thread to the next agent. Findings carry a scope, so recall answers the question that was asked rather than the nearest-sounding one.
Module 04: Blast radius and the impact envelope
Measure reach with nool debug blast-radius, watch a proposal refused for exceeding its declared scope, and land a justified change with --full validation. Scope creep becomes a gated, recorded event instead of a surprise in review.
Module 05: Parallel agents without collisions
Give two agents isolated worktrees and fenced leases with nool try new --worktree --nodes, see a second lease on a held scope fail with exit 3, check nool discover conflicts, and promote one branch at a time. Promote performs a real three-way merge and stops with diff3 markers on a same-symbol conflict.
Module 06: Tasks with acceptance criteria and rails
Create or import tasks with verifiable criteria, claim and start them, and verify them done against the knot that landed. Imported GitHub or Jira issues arrive with draft criteria for review, and each task links to the knots that close it. Rails name the paths a task's work may not touch, and a proposal that crosses one is refused until someone else signs off.
Module 07: Gates, policies and human holds
Scaffold governance with nool config init-governance and land it as its own configuration knot, run invariants with nool verify --all, resolve a steering checkpoint, record a signed attestation, then check the record control by control with nool audit verify and see which gates send work back with nool policy stats. Covers role grants, pre-tool checks on agent edits and policies that wait for evidence before a push.
Module 08: Fleets, reviewer quorums and souls
Get a fleet shape from nool fleet advise, preview waves with nool fleet plan, run a builder against a cross-vendor reviewer quorum, create a budget-capped persona and grade its run with nool eval. Quorums are cross-vendor on purpose, so no model reviews its own work.
Module 09: Debug, observe and keep the ledger healthy
Replay a change, walk from a failure to its cause, bisect a regression with a test command, read spend and waste with nool usage, and release with nool doctor --strict and nool checkpoint. These are the commands an operator runs on a schedule to keep the ledger trustworthy.
Module 10: Run a goal with nool code
Set up a model provider, trust the repository, land [code] checks and a minimum verdict, tighten permissions, and run one goal with a playbook and a task. Nool's completion check, not the model, decides when the goal is done, and grades each pass verified, self_checked or unverified. Then run a goal headless with --exec, branch on its exit code, and resume a session. Export a session as a diagram, and land each verified goal on its own branch with the pull request opened for you.
Module 11: Run and steer a fleet across machines
Plan a goal into tasks with nool code --fleet, watch it with nool fleet status, retry or cancel one task while the rest carry on, add Modal or Daytona sandboxes as pools mid-run and drain your own machine. Cloud agents never receive your model key: a gateway on your machine gives each one a capped token.
Module 12: Branches, conventions and merging
Declare branch naming conventions in nool.toml, create convention-named branches, see the git guard refuse a name that fits none, land a fleet on a feature branch, check commit messages against the Enterprise commit template, and merge the branch back with nool pr merge so both sides' history is recorded.
Module 13: Capstone: ship a feature with three agents
Combine every module: one task with three criteria, three agents in isolated worktrees, the gate stack, a reviewer quorum, CI attestation, task verification and a release. The course is complete when the audit report shows gates, approvals and council evidence for every knot in the feature.
Frequently asked questions
- Do I need a real project to follow the course?
- Any Git repository you can break safely works. The examples describe a webhook service with an API and a billing module, but every command applies to your own code.
- How long does it take?
- About four and a quarter hours end to end, with the capstone taking the longest. Modules are 10 to 25 minutes each and can be done across several sessions; progress is kept in your browser when preference cookies are accepted.
- Which agents can I use?
- Any agent that can run a CLI or call the Nool MCP server: Claude Code, Codex, Cursor, Gemini CLI or your own. Module 10 uses Nool's own agent, nool code, with a model key from any supported provider. Modules 05 and 08 need at least two agents or two sessions.
- Is the course free?
- Mostly. Community is free forever for one engineer with 1,000 knots a month, which covers the single-agent modules. Fleets in Modules 08 and 11 need Team, which you can try free for 30 days, and the commit template in Module 12 is an Enterprise feature.