diff --git a/CHANGELOG.md b/CHANGELOG.md
index 59713cb..b9eac61 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -69,3 +69,10 @@ Initial implementation.
 - Interactive screens now gate on a real terminal check (ioctl) instead of
   "is a character device": `traust-cli skills < /dev/null` used to open the
   picker and hang forever, because /dev/null is a character device.
+- Cobra help output respects the terminal width: flag usages wrap with
+  continuations aligned under each description (FlagUsagesWrapped at the
+  measured width), Long descriptions re-wrap, the trailing use-hint breaks
+  on narrow terminals, cobra's built-in completion commands are hidden from
+  the command list (still invocable), over-long command summaries were
+  shortened, and the init --forge default URL moved out of the flag line
+  (an unbreakable token that could not fit pflag's description column).
diff --git a/cmd/env.go b/cmd/env.go
index de564a2..19d9ad2 100644
--- a/cmd/env.go
+++ b/cmd/env.go
@@ -12,7 +12,7 @@ var envJSON bool
 
 var envCmd = &cobra.Command{
 	Use:   "env",
-	Short: "Print the environment the CLI injects into child processes",
+	Short: "Print the injected child environment",
 	Long: `The CLI does not write to your shell profile. It injects the harness
 venv PATH, TRAUST_CONFIG_HOME and HARNESS_BIN_DIR into every process it
 spawns. This command prints that environment — source it if you want the
diff --git a/cmd/init.go b/cmd/init.go
index 823cf38..2057ae1 100644
--- a/cmd/init.go
+++ b/cmd/init.go
@@ -46,7 +46,7 @@ TRAUST_CONFIG_HOME and the venv PATH into the processes it spawns. Pass
 func init() {
 	rootCmd.AddCommand(initCmd)
 	initCmd.Flags().StringVar(&initOpts.workspace, "workspace", "", "managed workspace root (default ~/.traust)")
-	initCmd.Flags().StringVar(&initOpts.forge, "forge", bootstrap.DefaultForge, "git URL to clone traust from")
+	initCmd.Flags().StringVar(&initOpts.forge, "forge", "", "git URL to clone traust from (default: the public traust repository)")
 	initCmd.Flags().StringVar(&initOpts.ref, "ref", "latest", "tag or branch to pin (or \"latest\" for the newest stable tag)")
 	initCmd.Flags().StringVar(&initOpts.configHome, "config-home", "", "TRAUST_CONFIG_HOME (default <workspace>/config)")
 	initCmd.Flags().StringVar(&initOpts.toolchain, "toolchain", "none", "scanner toolchain profile (none, secure-code-audit, vuln-scan, triage, …)")
@@ -56,6 +56,14 @@ func init() {
 	initCmd.Flags().BoolVarP(&initOpts.yes, "yes", "y", false, "non-interactive: accept all defaults")
 }
 
+// defaultForge is the wizard's placeholder for the forge URL.
+func defaultForge() string {
+	if initOpts.forge != "" {
+		return initOpts.forge
+	}
+	return bootstrap.DefaultForge
+}
+
 func runInit(cmd *cobra.Command, args []string) {
 	ctx := context.Background()
 	styles := theme.Default()
@@ -79,7 +87,7 @@ func runInit(cmd *cobra.Command, args []string) {
 		if defaultHarness == "" {
 			defaultHarness = "claude"
 		}
-		answers, err := tui.RunInitWizard(workspace, initOpts.forge, detected, defaultHarness)
+		answers, err := tui.RunInitWizard(workspace, defaultForge(), detected, defaultHarness)
 		if err != nil {
 			fatalf(1, "init wizard: %v", err)
 		}
diff --git a/cmd/ledger.go b/cmd/ledger.go
index a3366b1..044903b 100644
--- a/cmd/ledger.go
+++ b/cmd/ledger.go
@@ -40,7 +40,7 @@ var ledgerWhoamiCmd = &cobra.Command{
 
 var ledgerFindingsCmd = &cobra.Command{
 	Use:   "findings [layer-id]",
-	Short: "List resolved finding dispositions (one layer or all)",
+	Short: "List resolved finding dispositions",
 	Run: func(cmd *cobra.Command, args []string) {
 		v := newLedgerViews()
 		layer := ""
diff --git a/cmd/root.go b/cmd/root.go
index b43a513..bf57e47 100644
--- a/cmd/root.go
+++ b/cmd/root.go
@@ -37,6 +37,7 @@ journals.`,
 
 // Execute runs the root command.
 func Execute() {
+	installWrappedUsageTemplate()
 	if err := rootCmd.Execute(); err != nil {
 		styles := theme.Default()
 		fmt.Fprintln(os.Stderr, styles.Err.Render("error: "+err.Error()))
@@ -44,6 +45,48 @@ func Execute() {
 	}
 }
 
+// installWrappedUsageTemplate makes cobra's help respect the terminal width.
+// The default template emits flag usages unwrapped, so long descriptions
+// hard-wrap at the terminal's right edge with continuations at column zero
+// while the flags themselves are indented. pflag already knows how to wrap
+// (FlagUsagesWrapped aligns continuations under each description); the
+// template is patched in place rather than copied, so it tracks cobra's
+// default. Long descriptions and the trailing use-hint are wrapped the same
+// way, and cobra's built-in completion commands are hidden (their summaries
+// are too long for narrow terminals and they remain invocable). Templates
+// inherit from the root, so every subcommand is covered.
+func installWrappedUsageTemplate() {
+	width := tui.TerminalWidth()
+	if width < 60 {
+		width = 60
+	}
+	tmpl := rootCmd.UsageTemplate()
+	patched := strings.ReplaceAll(tmpl,
+		".FlagUsages | trimTrailingWhitespaces",
+		fmt.Sprintf(".FlagUsagesWrapped %d | trimTrailingWhitespaces", width))
+	if width < 78 {
+		patched = strings.ReplaceAll(patched,
+			"for more information about a command.",
+			"for more\ninformation about a command.")
+	}
+	if patched != tmpl {
+		rootCmd.SetUsageTemplate(patched)
+	}
+	rootCmd.CompletionOptions.HiddenDefaultCmd = true
+	wrapCommandLongs(rootCmd, width)
+}
+
+// wrapCommandLongs re-wraps every command's Long description to the current
+// width; cobra renders them verbatim.
+func wrapCommandLongs(cmd *cobra.Command, width int) {
+	if cmd.Long != "" {
+		cmd.Long = tui.Wrap(cmd.Long, width)
+	}
+	for _, sub := range cmd.Commands() {
+		wrapCommandLongs(sub, width)
+	}
+}
+
 func init() {
 	rootCmd.SetVersionTemplate(theme.Default().App.Render(" traust-cli ") + " {{.Version}}\n")
 }
diff --git a/cmd/run.go b/cmd/run.go
index e2b66cc..6e96ae3 100644
--- a/cmd/run.go
+++ b/cmd/run.go
@@ -32,7 +32,7 @@ var runFlags struct {
 
 var runCmd = &cobra.Command{
 	Use:   "run <skill> [args…]",
-	Short: "Execute a skill headlessly through your agent harness",
+	Short: "Run a skill headlessly through your harness",
 	Long: `Composes the skill's SKILL.md into a prompt, resolves the model-floor
 default from the registry, and drives the harness of your choice in headless
 mode. The agent always runs from the harness root — never inside the audit