diff --git a/crates/mergify-cli/src/main.rs b/crates/mergify-cli/src/main.rs
index 53f47f99..12dca9ec 100644
--- a/crates/mergify-cli/src/main.rs
+++ b/crates/mergify-cli/src/main.rs
@@ -49,6 +49,18 @@ fn main() -> ExitCode {
         println!("✅");
     }
 
+    // Hidden flag — used by `mergify_cli/tests/queue/test_skill.py`
+    // (and any future cross-language test) to learn the set of
+    // commands the Rust binary handles natively without resorting
+    // to a hardcoded list that drifts. Format is one
+    // `<group> <subcommand>` pair per line.
+    if argv.first().is_some_and(|a| a == "--list-native-commands") {
+        for (group, sub) in NATIVE_COMMANDS {
+            println!("{group} {sub}");
+        }
+        return ExitCode::SUCCESS;
+    }
+
     if let Some(cmd) = detect_native(&argv) {
         return run_native(cmd);
     }
@@ -62,6 +74,23 @@ fn main() -> ExitCode {
     }
 }
 
+/// Single source of truth for the `(group, subcommand)` pairs the
+/// Rust binary handles natively. Used by [`looks_native`] for argv
+/// recognition and by the `--list-native-commands` hidden flag so
+/// out-of-process tests can discover the list without hard-coding
+/// it. Add new entries here when porting a command; the matching
+/// `clap` `Subcommands` variant is what actually wires it up.
+const NATIVE_COMMANDS: &[(&str, &str)] = &[
+    ("config", "validate"),
+    ("config", "simulate"),
+    ("ci", "scopes-send"),
+    ("ci", "git-refs"),
+    ("ci", "queue-info"),
+    ("queue", "pause"),
+    ("queue", "unpause"),
+    ("queue", "status"),
+];
+
 /// Native commands the Rust binary handles without delegating to
 /// the Python shim.
 enum NativeCommand {
@@ -126,12 +155,9 @@ struct QueueStatusOpts {
 /// classify the invocation as native.
 fn looks_native(argv: &[String]) -> bool {
     argv.windows(2).any(|pair| {
-        matches!(
-            (pair[0].as_str(), pair[1].as_str()),
-            ("config", "validate" | "simulate")
-                | ("ci", "scopes-send" | "git-refs" | "queue-info")
-                | ("queue", "pause" | "unpause" | "status"),
-        )
+        NATIVE_COMMANDS
+            .iter()
+            .any(|(g, s)| pair[0] == *g && pair[1] == *s)
     })
 }
 
diff --git a/mergify_cli/tests/queue/test_skill.py b/mergify_cli/tests/queue/test_skill.py
index e701b83b..bbc98c07 100644
--- a/mergify_cli/tests/queue/test_skill.py
+++ b/mergify_cli/tests/queue/test_skill.py
@@ -16,6 +16,8 @@
 
 import pathlib
 import re
+import shutil
+import subprocess
 
 import yaml
 
@@ -29,6 +31,42 @@ def _get_skill_content() -> str:
     return SKILL_PATH.read_text(encoding="utf-8")
 
 
+def _native_commands_for_group(group: str) -> set[str]:
+    """Ask the installed `mergify` binary which `<group> <sub>` pairs
+    it handles natively, then return the subcommands for `group`.
+
+    Spawning the binary keeps this test honest: the source of truth
+    is the binary itself, so a port that adds a native subcommand
+    (and its `NATIVE_COMMANDS` entry) automatically becomes visible
+    here. No parallel hard-coded list to drift.
+    """
+    # Fail (don't skip) when `mergify` isn't on PATH: the wheel must
+    # be installed for the test to be meaningful, and silently
+    # skipping would let a regression in the wheel's bin scripts
+    # slip through unnoticed. `uv run pytest` installs the wheel
+    # before invoking pytest, so on every supported entry point the
+    # binary is present.
+    binary = shutil.which("mergify")
+    assert binary is not None, (
+        "`mergify` binary not on PATH. Install the wheel first "
+        "(`uv run pytest` does this automatically) — running this "
+        "test against an uninstalled checkout would give a false pass."
+    )
+    out = subprocess.run(
+        [binary, "--list-native-commands"],
+        check=True,
+        capture_output=True,
+        text=True,
+    ).stdout
+    pairs = (line.split(maxsplit=1) for line in out.splitlines() if line.strip())
+    return {
+        sub
+        for pair in pairs
+        if len(pair) == 2 and pair[0] == group
+        for sub in [pair[1]]
+    }
+
+
 def test_skill_content_is_readable() -> None:
     content = _get_skill_content()
     assert len(content) > 0
@@ -62,25 +100,23 @@ def test_skill_has_required_sections() -> None:
         assert section in content, f"Skill is missing required section: {section}"
 
 
-# Rust-native queue commands. Each port PR appends to this list when
-# it deletes the Python copy, so the validation below stays accurate
-# without needing to spawn the Rust binary at test time.
-NATIVE_QUEUE_COMMANDS: frozenset[str] = frozenset({"pause", "unpause", "status"})
-
-
 def test_skill_references_valid_commands() -> None:
     """Every `mergify queue <cmd>` reference in the skill must resolve
-    to either a registered click command (still-shimmed) or a known
-    Rust-native command. Catches typos and skill drift after a port."""
+    to either a registered click command (still-shimmed) or a
+    Rust-native command reported by the binary. Catches typos and
+    skill drift after a port — and stays accurate without a parallel
+    hard-coded list because the native set is queried from
+    `mergify --list-native-commands` itself.
+    """
     from mergify_cli.queue.cli import queue
 
     content = _get_skill_content()
     referenced = set(re.findall(r"mergify queue ([\w-]+)", content))
-    available = set(queue.commands.keys()) | NATIVE_QUEUE_COMMANDS
+    available = set(queue.commands.keys()) | _native_commands_for_group("queue")
 
     for cmd in referenced:
         assert cmd in available, (
             f"Skill references 'mergify queue {cmd}' but it's neither a "
-            f"registered click command nor a known Rust-native command. "
+            f"registered click command nor a Rust-native command. "
             f"Available: {sorted(available)}"
         )