diff --git a/tests/unit/test_cli/test_diagnose.py b/tests/unit/test_cli/test_diagnose.py
new file mode 100644
index 0000000..6a0fa05
--- /dev/null
+++ b/tests/unit/test_cli/test_diagnose.py
@@ -0,0 +1,278 @@
+"""Tests for the oaeval diagnose command."""
+
+from __future__ import annotations
+
+import json
+import re
+from typing import TYPE_CHECKING
+from unittest.mock import MagicMock
+
+import pytest
+from typer.testing import CliRunner
+
+from openagent_eval.cli.context import reset_context
+from openagent_eval.cli.main import app
+from openagent_eval.diagnosis.models import (
+    BlameTarget,
+    DiagnosisReport,
+    FailureInstance,
+    FailureMode,
+)
+
+if TYPE_CHECKING:
+    from pathlib import Path
+
+runner = CliRunner()
+
+
+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)
+
+
+@pytest.fixture(autouse=True)
+def _reset_cli_context() -> None:
+    """Reset the global CLI context before and after every test.
+
+    The CLI context (quiet/json/verbose flags) lives in a module-level
+    global, so leaving it dirty between tests can make later tests pass
+    or fail depending on run order.
+    """
+    reset_context()
+    yield
+    reset_context()
+
+
+@pytest.fixture
+def fake_analyzer(monkeypatch: pytest.MonkeyPatch) -> dict:
+    """Replace DiagnosisAnalyzer so no real model or analysis code is executed.
+
+    Yields a holder dict recording every constructed analyzer instance, its
+    constructor kwargs, and the DiagnosisReport returned from ``analyze``.
+    """
+    report = DiagnosisReport(
+        total_items=2,
+        blame_summary={BlameTarget.RETRIEVAL.value: 1},
+        failure_summary={FailureMode.EMPTY_RETRIEVAL.value: 1},
+        failures=[
+            FailureInstance(
+                mode=FailureMode.EMPTY_RETRIEVAL,
+                blame=BlameTarget.RETRIEVAL,
+                confidence=0.85,
+                reason="No contexts were retrieved.",
+                question="What is RAG?",
+                evidence={"context_count": 0},
+            ),
+        ],
+        chunking_issues=[],
+        recommendations=["Check if the retriever is using the correct embedding model."],
+        overall_health=0.5,
+    )
+    holder: dict = {"instances": [], "report": report}
+
+    def factory(**kwargs: object) -> MagicMock:
+        instance = MagicMock()
+        instance.init_kwargs = kwargs
+        instance.analyze = MagicMock(return_value=holder["report"])
+        holder["instances"].append(instance)
+        return instance
+
+    monkeypatch.setattr(
+        "openagent_eval.cli.commands.diagnose.DiagnosisAnalyzer", factory
+    )
+    return holder
+
+
+def _write_report(path: Path, payload: object) -> None:
+    """Write ``payload`` as JSON to ``path``."""
+    path.write_text(json.dumps(payload), encoding="utf-8")
+
+
+# --------------------------------------------------------------------------- #
+# Argument parsing and help
+# --------------------------------------------------------------------------- #
+
+
+def test_diagnose_help_lists_documented_flags() -> None:
+    """Help output exposes the command's public arguments and options."""
+    result = runner.invoke(app, ["diagnose", "--help"])
+    assert result.exit_code == 0
+    output = strip_ansi(result.output)
+    assert "report_path" in output
+    for flag in ("--output", "--threshold", "--max-recs", "--verbose"):
+        assert flag in output
+
+
+def test_diagnose_missing_argument_exits_two() -> None:
+    """Invoking diagnose without a report path is a usage error (exit 2)."""
+    result = runner.invoke(app, ["diagnose"])
+    assert result.exit_code == 2
+    assert "REPORT_PATH" in result.output or "Missing" in result.output
+
+
+# --------------------------------------------------------------------------- #
+# Input validation errors (exit code 2)
+# --------------------------------------------------------------------------- #
+
+
+def test_diagnose_missing_file_exits_two(tmp_path: Path) -> None:
+    """A non-existent report file surfaces as exit code 2."""
+    missing = tmp_path / "missing.json"
+    result = runner.invoke(app, ["diagnose", str(missing)])
+    assert result.exit_code == 2
+    assert "not found" in strip_ansi(result.output).lower()
+
+
+def test_diagnose_non_json_file_exits_two(tmp_path: Path) -> None:
+    """A report file that is not JSON surfaces as exit code 2."""
+    report = tmp_path / "report.txt"
+    report.write_text("not json", encoding="utf-8")
+    result = runner.invoke(app, ["diagnose", str(report)])
+    assert result.exit_code == 2
+    assert "json" in strip_ansi(result.output).lower()
+
+
+def test_diagnose_invalid_json_exits_two(tmp_path: Path) -> None:
+    """Malformed JSON in the report file surfaces as exit code 2."""
+    report = tmp_path / "report.json"
+    report.write_text("{not valid json", encoding="utf-8")
+    result = runner.invoke(app, ["diagnose", str(report)])
+    assert result.exit_code == 2
+    assert "parse" in strip_ansi(result.output).lower()
+
+
+# --------------------------------------------------------------------------- #
+# Report-loading edge cases
+# --------------------------------------------------------------------------- #
+
+
+def test_diagnose_empty_results_exits_zero(tmp_path: Path) -> None:
+    """A report containing an empty list of results exits cleanly with a warning."""
+    report = tmp_path / "report.json"
+    _write_report(report, [])
+    result = runner.invoke(app, ["diagnose", str(report)])
+    assert result.exit_code == 0, result.output
+    assert "no evaluation results" in strip_ansi(result.output).lower()
+
+
+def test_diagnose_unsupported_report_format_exits_zero(
+    tmp_path: Path,
+    fake_analyzer: dict,
+) -> None:
+    """A JSON payload that is neither a list nor a {results: ...} wrapper yields no results."""
+    report = tmp_path / "report.json"
+    _write_report(report, {"meta": "data"})
+    result = runner.invoke(app, ["diagnose", str(report)])
+    assert result.exit_code == 0, result.output
+    assert "no evaluation results" in strip_ansi(result.output).lower()
+    assert fake_analyzer["instances"] == []
+
+
+def test_diagnose_pipeline_result_wrapper_is_loaded(
+    tmp_path: Path,
+    fake_analyzer: dict,
+) -> None:
+    """A PipelineResult-style wrapper with a 'results' key is unpacked correctly."""
+    report = tmp_path / "report.json"
+    results = [{"question": "Q1", "answer": "A1"}]
+    _write_report(report, {"results": results})
+    result = runner.invoke(app, ["diagnose", str(report)])
+    assert result.exit_code == 0, result.output
+    assert len(fake_analyzer["instances"]) == 1
+    assert fake_analyzer["instances"][0].analyze.call_args.args[0] == results
+
+
+# --------------------------------------------------------------------------- #
+# Happy path and output formats
+# --------------------------------------------------------------------------- #
+
+
+def test_diagnose_happy_path_terminal_output(
+    tmp_path: Path,
+    fake_analyzer: dict,
+) -> None:
+    """The default terminal output renders the diagnosis report."""
+    report = tmp_path / "report.json"
+    _write_report(report, [{"question": "Q1", "answer": "A1"}])
+    result = runner.invoke(app, ["diagnose", str(report)])
+    assert result.exit_code == 0, result.output
+    output = strip_ansi(result.output)
+    assert "Diagnosis Report" in output
+    assert "Items analyzed: 2" in output
+    assert "System Health" in output
+    assert "Retrieval" in output
+    assert "Check if the retriever is using the correct embedding model" in output
+
+
+def test_diagnose_json_output_format(
+    tmp_path: Path,
+    fake_analyzer: dict,
+) -> None:
+    """--output json emits a JSON-serialized DiagnosisReport."""
+    report = tmp_path / "report.json"
+    _write_report(report, [{"question": "Q1", "answer": "A1"}])
+    result = runner.invoke(app, ["diagnose", str(report), "--output", "json"])
+    assert result.exit_code == 0, result.output
+    payload = json.loads(strip_ansi(result.output))
+    assert payload["total_items"] == 2
+    assert payload["blame_summary"] == {BlameTarget.RETRIEVAL.value: 1}
+    assert payload["overall_health"] == 0.5
+
+
+def test_diagnose_global_json_flag_triggers_json_output(
+    tmp_path: Path,
+    fake_analyzer: dict,
+) -> None:
+    """The global --json flag forces JSON output even when --output is omitted."""
+    report = tmp_path / "report.json"
+    _write_report(report, [{"question": "Q1", "answer": "A1"}])
+    result = runner.invoke(app, ["--json", "diagnose", str(report)])
+    assert result.exit_code == 0, result.output
+    payload = json.loads(strip_ansi(result.output))
+    assert payload["total_items"] == 2
+
+
+# --------------------------------------------------------------------------- #
+# Option wiring
+# --------------------------------------------------------------------------- #
+
+
+def test_diagnose_threshold_and_max_recs_passed_to_analyzer(
+    tmp_path: Path,
+    fake_analyzer: dict,
+) -> None:
+    """--threshold and --max-recs reach the DiagnosisAnalyzer constructor."""
+    report = tmp_path / "report.json"
+    _write_report(report, [{"question": "Q1", "answer": "A1"}])
+    result = runner.invoke(
+        app,
+        [
+            "diagnose",
+            str(report),
+            "--threshold",
+            "0.7",
+            "--max-recs",
+            "3",
+        ],
+    )
+    assert result.exit_code == 0, result.output
+    kwargs = fake_analyzer["instances"][0].init_kwargs
+    assert kwargs["blame_threshold"] == 0.7
+    assert kwargs["max_recommendations"] == 3
+
+
+def test_diagnose_verbose_shows_detailed_failures(
+    tmp_path: Path,
+    fake_analyzer: dict,
+) -> None:
+    """--verbose includes detailed failure information in terminal output."""
+    report = tmp_path / "report.json"
+    _write_report(report, [{"question": "Q1", "answer": "A1"}])
+    result = runner.invoke(app, ["diagnose", str(report), "--verbose"])
+    assert result.exit_code == 0, result.output
+    output = strip_ansi(result.output)
+    assert "Detailed Failures" in output
+    assert "Empty Retrieval" in output
+    assert "No contexts were retrieved" in output
+    assert "What is RAG?" in output
diff --git a/tests/unit/test_corpus/test_auditor.py b/tests/unit/test_corpus/test_auditor.py
index d9b6ca4..4a55fbd 100644
--- a/tests/unit/test_corpus/test_auditor.py
+++ b/tests/unit/test_corpus/test_auditor.py
@@ -110,8 +110,22 @@ def test_build_analyzers_rejects_unknown_check_names(self):
     @pytest.mark.asyncio
     async def test_audit_with_valid_checks_still_works(self, temp_corpus):
         """A fully valid check list still produces a normal audit report."""
+        import sys
+        from unittest.mock import MagicMock, patch
+
+        import numpy as np
+
+        # Stub the embedding model so the duplicate analyzer runs its real
+        # similarity pipeline without downloading weights from the Hugging
+        # Face hub — network access is not guaranteed in CI.
+        mock_st = MagicMock()
+        mock_st.SentenceTransformer.return_value.encode.return_value = np.array(
+            [[1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 1.0]]
+        )
+
         auditor = CorpusAuditor(checks=["staleness", "duplicate"])
-        report = await auditor.audit(str(temp_corpus))
+        with patch.dict(sys.modules, {"sentence_transformers": mock_st}):
+            report = await auditor.audit(str(temp_corpus))
 
         assert report.total_documents == 3
         assert "staleness" in report.checks_performed