diff --git a/src/runtime/cli/mod.rs b/src/runtime/cli/mod.rs
index 9bd01c88ea88..47db997bf0b7 100644
--- a/src/runtime/cli/mod.rs
+++ b/src/runtime/cli/mod.rs
@@ -890,6 +890,14 @@ pub mod command {
         strings::contains_char(basename, b'.')
     }
 
+    /// `-e`/`-p` take a value, so any `-e…`/`-p…` token is one; clusters like `-bp` are not.
+    #[inline]
+    fn is_eval_flag(arg: &[u8]) -> bool {
+        matches!(arg, [b'-', b'e' | b'p', ..] | b"--eval" | b"--print")
+            || arg.starts_with(b"--eval=")
+            || arg.starts_with(b"--print=")
+    }
+
     /// `#[inline(never)]`: argv→`Tag` classification, called once from
     /// `Cli::start` on every `bun` invocation. Kept a concrete symbol so
     /// `src/startup.order` can place it next to `Cli::start` /
@@ -937,10 +945,11 @@ pub mod command {
         let Some(mut first_arg_name) = iter.next() else {
             return Tag::AutoCommand;
         };
-        while !first_arg_name.is_empty()
-            && first_arg_name[0] == b'-'
-            && !(first_arg_name.len() > 1 && first_arg_name[1] == b'e')
-        {
+        while !first_arg_name.is_empty() && first_arg_name[0] == b'-' {
+            // The rest of argv belongs to the script: https://github.com/oven-sh/bun/issues/23631
+            if is_eval_flag(first_arg_name) {
+                return Tag::AutoCommand;
+            }
             // `--interactive` stays on AutoCommand: Arguments.rs parses it and the no-target check
             // routes to RunCommand::exec_node_repl. An early ReplCommand return here would bypass
             // that and boot the legacy `bun repl` implementation instead.
@@ -1068,9 +1077,6 @@ pub mod command {
             }
             return Tag::AutoCommand;
         }
-        if x == RootCommandMatcher::case(b"-e") {
-            return Tag::AutoCommand;
-        }
         Tag::AutoCommand
     }
 
diff --git a/test/cli/run/run-eval.test.ts b/test/cli/run/run-eval.test.ts
index 0f928a7ce6a7..b9fca9710a34 100644
--- a/test/cli/run/run-eval.test.ts
+++ b/test/cli/run/run-eval.test.ts
@@ -1,7 +1,7 @@
 import { SyncSubprocess } from "bun";
 import { describe, expect, test } from "bun:test";
 import { rmSync, writeFileSync } from "fs";
-import { bunEnv, bunExe, isWindows, tmpdirSync } from "harness";
+import { bunEnv, bunExe, isWindows, tempDir, tmpdirSync } from "harness";
 import { tmpdir } from "os";
 import { join, sep } from "path";
 
@@ -108,6 +108,70 @@ for (const flag of ["-e", "--print"]) {
   });
 }
 
+// https://github.com/oven-sh/bun/issues/23631
+// Empty cwd: a token dispatched as a subcommand (`bun test`, `bun x`, ...) must find nothing to act on.
+describe("eval flags take precedence over subcommand names", () => {
+  describe("the code is a subcommand name", () => {
+    const cases: string[][] = [
+      ["-p", "test"],
+      ["--print", "x"],
+      ["--eval", "test"],
+      ["-pe", "help"],
+      ["-e", "test"],
+      // The eval flag is reached after skipping an unrelated flag.
+      ["--smol", "-p", "test"],
+    ];
+
+    for (const args of cases) {
+      const name = args[args.length - 1];
+      test.concurrent(`bun ${args.join(" ")} evaluates \`${name}\``, async () => {
+        using dir = tempDir("eval-subcommand-name", {});
+        await using proc = Bun.spawn({
+          cmd: [bunExe(), ...args],
+          cwd: String(dir),
+          env: bunEnv,
+          stdout: "pipe",
+          stderr: "pipe",
+        });
+        const [stdout, stderr, exitCode] = await Promise.all([proc.stdout.text(), proc.stderr.text(), proc.exited]);
+        expect(stderr).toContain(`ReferenceError: ${name} is not defined`);
+        expect(stdout).toBe("");
+        expect(exitCode).toBe(1);
+      });
+    }
+  });
+
+  describe("a script argument is a subcommand name", () => {
+    const argv = "JSON.stringify(process.argv.slice(1))";
+    const cases: string[][] = [
+      [`-p${argv}`, "test"],
+      [`-p=${argv}`, "test"],
+      [`--print=${argv}`, "test"],
+      [`--eval=console.log(${argv})`, "help"],
+      [`-e=console.log(${argv})`, "test"],
+      ["-p", argv, "test"],
+    ];
+
+    for (const args of cases) {
+      const name = args[args.length - 1];
+      test.concurrent(`bun ${args.join(" ")} passes \`${name}\` to the script`, async () => {
+        using dir = tempDir("eval-subcommand-arg", {});
+        await using proc = Bun.spawn({
+          cmd: [bunExe(), ...args],
+          cwd: String(dir),
+          env: bunEnv,
+          stdout: "pipe",
+          stderr: "pipe",
+        });
+        const [stdout, stderr, exitCode] = await Promise.all([proc.stdout.text(), proc.stderr.text(), proc.exited]);
+        expect(stderr).toBe("");
+        expect(stdout).toBe(JSON.stringify([name]) + "\n");
+        expect(exitCode).toBe(0);
+      });
+    }
+  });
+});
+
 describe("--print for cjs/esm", () => {
   test("eval result between esm imports", async () => {
     let cwd = tmpdirSync();