Add consistent --version flag to all CLI commands (#125)

CHANGELOG.md

@@ -8,7 +8,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 
 ## [Unreleased]
 
-No changes yet.
+### Fixed
+
+- Add consistent `--version` flag to all CLI commands (#125).
 
 ---
 

openagent_eval/cli/main.py

@@ -3,11 +3,14 @@
 from __future__ import annotations
 
 import sys
+from importlib.metadata import PackageNotFoundError, version
 
 import typer
 from rich.console import Console
+from typer.core import TyperGroup
 
 from openagent_eval.cli.banner import create_mini_banner
+from openagent_eval.cli.commands.audit import audit_command
 from openagent_eval.cli.commands.compare import compare_command
 from openagent_eval.cli.commands.delete import delete_command
 from openagent_eval.cli.commands.diagnose import diagnose_command
@@ -19,9 +22,7 @@
 from openagent_eval.cli.commands.synth import synth_command
 from openagent_eval.cli.commands.test import test_command
 from openagent_eval.cli.commands.validate import validate_command
-from openagent_eval.cli.commands.audit import audit_command
 from openagent_eval.cli.context import CLIContext, set_context
-from openagent_eval.cli.utils.callbacks import version_callback
 from openagent_eval.exceptions import OpenAgentEvalError
 
 # Error code mapping for different exception types
@@ -33,7 +34,31 @@
     "CorpusError": 6,
 }
 
+
+def version_callback(value: bool) -> None:
+    if value:
+        try:
+            v = version("openagent-eval")
+        except PackageNotFoundError:
+            v = "unknown"
+        typer.echo(f"openagent-eval {v}")
+        raise typer.Exit()
+
+
+class VersionAwareGroup(TyperGroup):
+    def parse_args(self, ctx, args):
+        if args:
+            version_flags = {"--version", "-V"}
+            for flag in version_flags:
+                if flag in args:
+                    # Reconstruct args with the version flag at the beginning
+                    args = [flag] + [arg for arg in args if arg != flag]
+                    break
+        return super().parse_args(ctx, args)
+
+
 app = typer.Typer(
+    cls=VersionAwareGroup,
     name="oaeval",
     help="Open-source CLI framework for evaluating RAG systems and AI Agents.",
     no_args_is_help=True,
@@ -45,10 +70,10 @@
 @app.callback(invoke_without_command=True)
 def main(
     version: bool = typer.Option(
-        False,
+        None,
         "--version",
         "-V",
-        help="Show version and exit.",
+        help="Show the application's version and exit.",
         callback=version_callback,
         is_eager=True,
     ),
@@ -454,7 +479,9 @@ def _generate_fish_completion() -> str:
 
 
 # Override the default exception handler
-def _cli_exception_handler(exc_type: type, exc_value: BaseException, exc_tb: object) -> None:
+def _cli_exception_handler(
+    exc_type: type, exc_value: BaseException, exc_tb: object
+) -> None:
     """Custom exception handler for CLI.
 
     Args:

tests/unit/test_cli/test_cli_improvements.py

@@ -23,8 +23,8 @@
 
 def strip_ansi(text: str) -> str:
     """Strip ANSI escape codes from text."""
-    ansi_escape = re.compile(r'\x1b\[[0-9;]*m')
-    return ansi_escape.sub('', text)
+    ansi_escape = re.compile(r"\x1b\[[0-9;]*m")
+    return ansi_escape.sub("", text)
 
 
 class TestGlobalFlags:
@@ -120,7 +120,11 @@ def test_delete_nonexistent_report(self):
         """Test delete with nonexistent report ID."""
         result = runner.invoke(app, ["delete", "nonexistent_report"])
         # Should fail or show error
-        assert result.exit_code != 0 or "Error" in result.output or "not found" in result.output
+        assert (
+            result.exit_code != 0
+            or "Error" in result.output
+            or "not found" in result.output
+        )
 
 
 class TestCompletionCommand:
@@ -252,7 +256,9 @@ def test_find_config_file_found(self, tmp_path):
         from openagent_eval.cli.utils.discovery import find_config_file
 
         config_file = tmp_path / "config.yaml"
-        config_file.write_text("dataset: test.yaml\nllm:\n  provider: openai\n  model: gpt-4")
+        config_file.write_text(
+            "dataset: test.yaml\nllm:\n  provider: openai\n  model: gpt-4"
+        )
         result = find_config_file(tmp_path)
         assert result == config_file
 
@@ -261,7 +267,9 @@ def test_find_config_file_oaeval_name(self, tmp_path):
         from openagent_eval.cli.utils.discovery import find_config_file
 
         config_file = tmp_path / "oaeval.yaml"
-        config_file.write_text("dataset: test.yaml\nllm:\n  provider: openai\n  model: gpt-4")
+        config_file.write_text(
+            "dataset: test.yaml\nllm:\n  provider: openai\n  model: gpt-4"
+        )
         result = find_config_file(tmp_path)
         assert result == config_file
 
@@ -271,13 +279,33 @@ class TestVersionCommand:
 
     def test_version_short_flag(self):
         """Test --version flag."""
+        from importlib.metadata import version
+
         result = runner.invoke(app, ["--version"])
         assert result.exit_code == 0
+        assert result.output.strip() == f"openagent-eval {version('openagent-eval')}"
 
-    def test_version_V_flag(self):
+    def test_version_V_flag(self):  # noqa: N802
         """Test -V flag."""
+        from importlib.metadata import version
+
         result = runner.invoke(app, ["-V"])
         assert result.exit_code == 0
+        assert result.output.strip() == f"openagent-eval {version('openagent-eval')}"
+
+    def test_version_on_subcommand(self):
+        """Test --version flag on subcommands."""
+        from importlib.metadata import version
+
+        expected = f"openagent-eval {version('openagent-eval')}"
+
+        result_run = runner.invoke(app, ["run", "--version"])
+        assert result_run.exit_code == 0
+        assert result_run.output.strip() == expected
+
+        result_init = runner.invoke(app, ["init", "-V"])
+        assert result_init.exit_code == 0
+        assert result_init.output.strip() == expected
 
 
 class TestHelpOutput: