diff --git a/CHANGELOG.md b/CHANGELOG.md
index 6da223c..6739737 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,21 @@ All notable changes to DirXMLDevWeb are documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project uses [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.74.0] - 2026-10-22
+
+Needs DirXMLDev 0.50.0.
+
+### Changed
+
+- **Tutorial: do it again with an agent** (docs/tutorial.md §9): make a token, connect Claude Code,
+  run the setup prompt, then have the agent make the change on a branch, plan and deploy it with
+  the environment's name typed by you, and find its lines in the audit log.
+
+### Fixed
+
+- The agent tokens dialog wraps the token, the command and the setup prompt instead of cutting
+  their long lines off.
+
 ## [0.73.0] - 2026-10-22
 
 Needs DirXMLDev 0.50.0.
diff --git a/docs/screenshots/tutorial/18-agent-tokens.png b/docs/screenshots/tutorial/18-agent-tokens.png
new file mode 100644
index 0000000..573f586
Binary files /dev/null and b/docs/screenshots/tutorial/18-agent-tokens.png differ
diff --git a/docs/tutorial.md b/docs/tutorial.md
index c053ee4..cfeb949 100644
--- a/docs/tutorial.md
+++ b/docs/tutorial.md
@@ -2,7 +2,9 @@
 
 This walk-through starts with nothing but Docker and an Identity Vault you can reach. It installs
 DirXMLDevWeb on your machine, reads a driver set from the vault into a git repository, changes one
-policy, commits the change and deploys it back to the vault. It takes about fifteen minutes.
+policy, commits the change and deploys it back to the vault. It takes about fifteen minutes. A last,
+optional section does it again with an agent, Claude Code, which works through the same server as
+you.
 
 Use a **lab or development vault** for this, never production: the last step writes to it. Pick a
 policy whose change cannot hurt anything. The screenshots use a lab vault with a scratch Loopback
@@ -209,6 +211,93 @@ Under the deploy's result, choose **Commit the deploy record**:
 It commits that file alone, as *Deployed to lab (Tutorial Loopback)*. Anything else you have not
 committed stays as it was. The bar then offers **Push**: choose it so that others see the record.
 
+## 9. Do it again with an agent
+
+An agent can do what you just did: change a policy, check it, commit it and deploy it. It acts as
+you, under your roles, and goes by the same rules as the page: it shows you each change before
+applying it, and it writes to the vault only after you have seen the plan and typed the
+environment's name. This section uses [Claude Code](https://claude.com/claude-code), installed on
+the same machine. Any agent that speaks MCP works the same way ([agents.md](agents.md)).
+
+### Make a token
+
+Choose **agent tokens…** beside your name, at the top right. Give the token a label,
+`claude-laptop`, keep **90 days**, and choose **Make the token**:
+
+![The agent tokens dialog: one token, claude-laptop, and how to set up your agent](screenshots/tutorial/18-agent-tokens.png)
+
+The token is shown once, above the list, with the command that connects Claude Code. Choose
+**Copy the command**. Treat the token like a password: whoever has it acts as you.
+
+### Connect Claude Code
+
+Paste the command into a terminal (PowerShell on Windows). It looks like this, with your token in
+it; `--scope user` makes the server available wherever you start Claude Code:
+
+```bash
+claude mcp add --scope user --transport http dirxmldev http://localhost:8080/mcp --header "Authorization: Bearer <your token>"
+```
+
+`claude mcp list` should show `dirxmldev` as connected.
+
+### Set it up
+
+Start `claude` and type:
+
+```
+/mcp__dirxmldev__setup
+```
+
+The server gives Claude Code its rules and two skills, one for editing and deploying a driver set
+and one for testing policies with the simulator. Claude Code installs the skills in
+`~/.claude/skills`, asking before it replaces one that's already there. It then tells you who it
+acts as (you, with your roles), the branch (`main`, the one you have open), the drivers, and what
+validation says. It changes nothing.
+
+Claude Code asks your permission the first time it uses each of the server's tools. Allow them as
+they come, or for the session.
+
+### Ask for the change
+
+This time, do the work on a branch of its own, as real work goes. Type:
+
+> On a new branch called tutorial/agent, change the Tutorial Loopback greeting to "Hello from my
+> agent". Show me the change before you apply it. Then validate, commit and push.
+
+The agent:
+
+1. opens a new branch, `tutorial/agent`, from `main`. The page stays on `main`: the agent's branch
+   is its own;
+2. reads the policy and shows you the change. Nothing is written yet: the server checked it as a
+   preview, the way **Review changes** does;
+3. applies it when you say yes, validates the tree, then commits and pushes.
+
+To see the agent's work on the page, choose **switch** beside the branch name and open
+`tutorial/agent`. The commit is there, under your name.
+
+### Deploy it
+
+Type:
+
+> Compare tutorial/agent with lab and plan the deploy.
+
+The agent shows the one difference, the changed policy, and the plan, as steps 1 and 2 of
+**Compare & deploy** do. To deploy, it needs the environment's name from you: type `lab` when it
+asks. It then deploys, with the same snapshot, write, driver restart and check as on the page, and
+tells you the result. Finish with:
+
+> Commit the deploy record and push.
+
+On `lab` a branch may be deployed. A production environment takes deploys from `main` only,
+pushed and current, and the server refuses anything else, from an agent as from you
+([git-workflow.md](git-workflow.md)).
+
+### What it did, as you
+
+Every call the agent made is in the audit log (`~/idmweb/data/audit.log`) as yours, marked
+`"via":"claude-laptop"`. When you no longer need the token, choose **revoke** beside it in the
+dialog; the agent is then refused at once.
+
 ## Where things are
 
 | What | Where |
@@ -218,7 +307,8 @@ committed stays as it was. The bar then offers **Push**: choose it so that other
 | Deploy snapshots | the workspace's `tree/deploy-snapshots/<environment>/`, never committed |
 | The deploy record | `tree/deploy-log/<environment>.jsonl`, committed |
 | The environments | `~/idmweb/config/environments.properties`; your bind DN and password in `~/idmweb/data/users/<you>/environments.properties` |
-| Who did what | `~/idmweb/data/audit.log` |
+| Who did what | `~/idmweb/data/audit.log`; an agent's lines say `via` its token |
+| Your agent tokens | `~/idmweb/data/users/<you>/agent-tokens.json`, a hash of each, never the token |
 
 ## If something went wrong
 
@@ -228,6 +318,8 @@ committed stays as it was. The bar then offers **Push**: choose it so that other
 - **Undo an edit:** **Discard** in the editor, before you apply. After you apply, and before you
   commit: open the uncommitted files from the bar and pick the file. **discard…** undoes the whole
   file; the arrow beside one changed block undoes just that block, then **Save**.
+- **The agent can't connect, or is refused with 401:** the token was mistyped, has expired or was
+  revoked. Make a new one, then `claude mcp remove dirxmldev` and add it again with the new command.
 - **The vault changed outside the tree** (someone used Designer or iManager): choose **Compare**,
   then **Update this branch from lab…** under the differences. It uses what Compare just read, so
   it opens at once, and shows what would change in your tree. Apply writes it there for you to
@@ -235,7 +327,7 @@ committed stays as it was. The bar then offers **Push**: choose it so that other
 
 ## Next: work with branches and environments
 
-You changed `main` directly and deployed to one vault. Real work goes differently: each change on
+In steps 5 to 8 you changed `main` directly and deployed to one vault. Real work goes differently: each change on
 a branch of its own, deployed to dev, merged into `main`, then promoted to test and production.
 [git-workflow.md](git-workflow.md) walks through that, and through setting up the environments of
 one driver set. Read it next.
@@ -247,8 +339,7 @@ Also:
   `docker exec idmweb idm vault.deploy tree --env lab --dry-run` plans a deploy.
 - Add the engine jars to the folder's `lib/` for the compile checks and the simulator
   ([quickstart.md](quickstart.md#4-optional-the-engine-jars)).
-- Bring your agent: make a token (your name at the top right → *agent tokens…*), connect Claude
-  Code with the command it shows, and run `/mcp__dirxmldev__setup`. It can then do what you did
-  here, asking you before it writes to the vault ([agents.md](agents.md#quick-start)).
+- [agents.md](agents.md) lists every tool an agent has, with its arguments, and how to connect
+  other agents.
 - [features.md](features.md) lists every view.
 - [hosted-install.md](hosted-install.md) is the same server for a team.
diff --git a/pom.xml b/pom.xml
index 76d0237..ff7e0b7 100644
--- a/pom.xml
+++ b/pom.xml
@@ -6,7 +6,7 @@
 
     <groupId>com.pointblue.dirxml</groupId>
     <artifactId>dirxml-dev-web</artifactId>
-    <version>0.73.0</version>
+    <version>0.74.0</version>
     <licenses>
         <license>
             <name>MIT License</name>
diff --git a/src/main/java/com/pointblue/dirxml/web/Serve.java b/src/main/java/com/pointblue/dirxml/web/Serve.java
index c6452a5..2be12d2 100644
--- a/src/main/java/com/pointblue/dirxml/web/Serve.java
+++ b/src/main/java/com/pointblue/dirxml/web/Serve.java
@@ -14,7 +14,7 @@
  */
 public final class Serve {
 
-    public static final String VERSION = "0.73.0";
+    public static final String VERSION = "0.74.0";
 
     private Serve() {
     }
diff --git a/web/src/components/AgentTokensDialog.tsx b/web/src/components/AgentTokensDialog.tsx
index 32beaef..d288c93 100644
--- a/web/src/components/AgentTokensDialog.tsx
+++ b/web/src/components/AgentTokensDialog.tsx
@@ -65,9 +65,9 @@ export function AgentTokensDialog({ onClose }: { onClose: () => void }) {
           {made?.token && (
             <div className="banner ok" data-testid="agent-token-made">
               <strong>&quot;{made.label}&quot; is made. Copy it now: it is not shown again.</strong>
-              <pre className="outcome small">{made.token}</pre>
+              <pre className="outcome small wrapped">{made.token}</pre>
               <p className="small">Connect Claude Code (keep the token out of files you commit):</p>
-              <pre className="outcome small">{command}</pre>
+              <pre className="outcome small wrapped">{command}</pre>
               <button onClick={() => void navigator.clipboard.writeText(command)}>Copy the command</button>
             </div>
           )}
@@ -131,7 +131,7 @@ export function AgentTokensDialog({ onClose }: { onClose: () => void }) {
             </li>
             <li>
               In Claude Code, run <code>/mcp__dirxmldev__setup</code>. In another agent, paste this; it holds no secret:
-              <pre className="outcome small" data-testid="agent-setup-prompt">{setupPrompt}</pre>
+              <pre className="outcome small wrapped" data-testid="agent-setup-prompt">{setupPrompt}</pre>
               <button onClick={() => void navigator.clipboard.writeText(setupPrompt)} disabled={!setupPrompt}>
                 Copy the prompt
               </button>
diff --git a/web/src/styles.css b/web/src/styles.css
index 8fc277f..b12c8ac 100644
--- a/web/src/styles.css
+++ b/web/src/styles.css
@@ -625,6 +625,11 @@ textarea {
   margin-top: 14px;
   max-width: 720px;
 }
+/* a token, a command or a prompt to copy: every character visible, no sideways scroll */
+pre.wrapped {
+  white-space: pre-wrap;
+  overflow-wrap: anywhere;
+}
 
 /* review dialog */
 .modal-back {