diff --git a/hive-bash-mcp/src/runner.rs b/hive-bash-mcp/src/runner.rs
index 424bb9f9..683cab11 100644
--- a/hive-bash-mcp/src/runner.rs
+++ b/hive-bash-mcp/src/runner.rs
@@ -526,15 +526,12 @@ async fn run_task(mut task: TaskFile, socket: &Path) {
     if let Err(e) = write_task(&task) {
         tracing::warn!(id = %id, error = ?e, "bash_runner: write running state failed");
     }
-    // No command text in the summary: `id` is already the human-chosen
-    // `name` when the caller passed one, and even the auto-generated
-    // hex id was already returned to the agent in the `run` tool's own
-    // result — the todo only needs to say which task, not repeat the
-    // command. An unbounded raw command (e.g. a multi-line heredoc
-    // script) would balloon every UI that renders todo summaries; the
-    // full command stays on disk in the task file for `status`/`view`
-    // instead.
-    upsert_bash_todo(socket, &id, format!("bash task `{id}` running")).await;
+    upsert_bash_todo(
+        socket,
+        &id,
+        format!("bash task `{id}` running: `{}`", task.cmd),
+    )
+    .await;
     // Best-effort: tally the normalised command head for the /stats
     // "favorite tools" view. Counted once per execution, regardless of
     // exit status. Never fails the task.