| """Test harness: chat, tools, skills, stats.""" |
|
|
| import sys |
| import os |
| sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..")) |
|
|
| from splitbit_llm.harness.harness import SplitBitHarness |
| from splitbit_llm.harness.tools import ToolRegistry, get_default_tools, tool_loop, parse_tool_calls |
|
|
|
|
| def test_harness_chat(): |
| """Test harness chat functionality.""" |
| harness = SplitBitHarness() |
| result = harness.chat("Hello there", channel="cli") |
|
|
| assert "response" in result, "No response in result" |
| assert "elapsed_s" in result, "No elapsed time" |
| assert "stats" in result, "No stats" |
| print(f" Response: {result['response'][:60]}") |
| print(f" Elapsed: {result['elapsed_s']}s") |
|
|
|
|
| def test_harness_stats(): |
| """Test harness stats aggregation.""" |
| harness = SplitBitHarness() |
| stats = harness.get_stats() |
|
|
| assert "model" in stats, "No model stats" |
| assert "skills" in stats, "No skills stats" |
| assert "recursive_links" in stats, "No link stats" |
| assert "auto_sizer" in stats, "No auto_sizer stats" |
| print(f" Model params: {stats['model']['param_count']:,}") |
| print(f" Tier: {stats['auto_sizer']['tier']}") |
|
|
|
|
| def test_tool_registry(): |
| """Test tool registry and execution.""" |
| registry = ToolRegistry() |
| for tool in get_default_tools(): |
| registry.register(tool) |
|
|
| assert len(registry.list_tools()) == 9, f"Wrong tool count: {len(registry.list_tools())}" |
| print(f" Tools: {len(registry.list_tools())}") |
|
|
| |
| result = registry.execute("calculate", "2 + 2") |
| assert result.success, f"Calculate failed: {result.error}" |
| assert "4" in result.output, f"Wrong result: {result.output}" |
| print(f" calculate(2+2) = {result.output}") |
|
|
|
|
| def test_tool_loop(): |
| """Test tool execution loop.""" |
| registry = ToolRegistry() |
| for tool in get_default_tools(): |
| registry.register(tool) |
|
|
| text = "Let me calculate: [TOOL: calculate(3 * 7)]" |
| final_text, results = tool_loop(text, registry) |
|
|
| assert len(results) == 1, f"Expected 1 result, got {len(results)}" |
| assert results[0].success, "Tool execution failed" |
| assert "21" in results[0].output, f"Wrong output: {results[0].output}" |
| print(f" Tool loop: {len(results)} calls, output: {results[0].output}") |
|
|
|
|
| def test_parse_tool_calls(): |
| """Test parsing tool calls from text.""" |
| text = "I will [TOOL: calculate(1 + 1)] and then [TOOL: calculate(2 + 2)]" |
| calls = parse_tool_calls(text) |
| assert len(calls) == 2, f"Expected 2 calls, got {len(calls)}" |
| assert calls[0][0] == "calculate", f"Wrong tool name: {calls[0][0]}" |
| print(f" Parsed {len(calls)} tool calls") |
|
|
|
|
| if __name__ == "__main__": |
| print("Running harness tests...") |
| test_harness_chat() |
| print(" ✓ test_harness_chat") |
| test_harness_stats() |
| print(" ✓ test_harness_stats") |
| test_tool_registry() |
| print(" ✓ test_tool_registry") |
| test_tool_loop() |
| print(" ✓ test_tool_loop") |
| test_parse_tool_calls() |
| print(" ✓ test_parse_tool_calls") |
| print("\nAll harness tests passed!") |
|
|
