| 1 | from __future__ import annotations |
| 2 | |
| 3 | import sys |
| 4 | from pathlib import Path |
| 5 | |
| 6 | import pytest |
| 7 | |
| 8 | PROJECT_ROOT = Path(__file__).resolve().parents[1] |
| 9 | if str(PROJECT_ROOT) not in sys.path: |
| 10 | sys.path.insert(0, str(PROJECT_ROOT)) |
| 11 | |
| 12 | from helpers.extract_tools import ( |
| 13 | extract_tool_request, |
| 14 | is_misformatted_tool_request, |
| 15 | json_parse_dirty, |
| 16 | normalize_tool_request, |
| 17 | ) |
| 18 | from helpers import parallel_tools |
| 19 | |
| 20 | |
| 21 | def test_normalize_tool_request_accepts_canonical_keys() -> None: |
| 22 | assert normalize_tool_request({"tool_name": "response", "tool_args": {"text": "ok"}}) == ( |
| 23 | "response", |
| 24 | {"text": "ok"}, |
| 25 | ) |
| 26 | |
| 27 | |
| 28 | def test_normalize_tool_request_accepts_fallback_keys() -> None: |
| 29 | assert normalize_tool_request({"tool": "response", "args": {"text": "ok"}}) == ( |
| 30 | "response", |
| 31 | {"text": "ok"}, |
| 32 | ) |
| 33 | |
| 34 | |
| 35 | def test_normalize_tool_request_uses_fallback_when_canonical_name_is_empty() -> None: |
| 36 | assert normalize_tool_request( |
| 37 | {"tool_name": "", "tool": "response", "args": {"text": "ok"}} |
| 38 | ) == ("response", {"text": "ok"}) |
| 39 | |
| 40 | |
| 41 | def test_normalize_tool_request_uses_fallback_when_canonical_args_are_invalid() -> None: |
| 42 | assert normalize_tool_request( |
| 43 | {"tool_name": "response", "tool_args": None, "args": {"text": "ok"}} |
| 44 | ) == ("response", {"text": "ok"}) |
| 45 | |
| 46 | |
| 47 | def test_normalize_tool_request_translates_method_suffix_to_action() -> None: |
| 48 | assert normalize_tool_request( |
| 49 | {"tool_name": "text_editor:read", "tool_args": {"path": "README.md"}} |
| 50 | ) == ("text_editor", {"path": "README.md", "action": "read"}) |
| 51 | |
| 52 | |
| 53 | def test_normalize_tool_request_translates_method_arg_to_action() -> None: |
| 54 | assert normalize_tool_request( |
| 55 | {"tool_name": "scheduler", "tool_args": {"method": "list_tasks"}} |
| 56 | ) == ("scheduler", {"method": "list_tasks", "action": "list_tasks"}) |
| 57 | |
| 58 | |
| 59 | def test_normalize_tool_request_preserves_explicit_action_over_method() -> None: |
| 60 | assert normalize_tool_request( |
| 61 | { |
| 62 | "tool_name": "scheduler:delete_task", |
| 63 | "tool_args": {"method": "list_tasks", "action": "show_task"}, |
| 64 | } |
| 65 | ) == ( |
| 66 | "scheduler", |
| 67 | {"method": "list_tasks", "action": "show_task"}, |
| 68 | ) |
| 69 | |
| 70 | |
| 71 | def test_normalize_tool_request_rejects_missing_args() -> None: |
| 72 | with pytest.raises(ValueError, match="tool_args"): |
| 73 | normalize_tool_request({"tool_name": "response"}) |
| 74 | |
| 75 | |
| 76 | def test_normalize_tool_request_accepts_native_function_format() -> None: |
| 77 | request = { |
| 78 | "type": "function", |
| 79 | "name": "search_engine", |
| 80 | "parameters": {"query": "latest Agent Zero release"}, |
| 81 | } |
| 82 | |
| 83 | assert json_parse_dirty(str(request)) == request |
| 84 | assert normalize_tool_request(request) == ( |
| 85 | "search_engine", |
| 86 | {"query": "latest Agent Zero release"}, |
| 87 | ) |
| 88 | |
| 89 | |
| 90 | def test_normalize_tool_request_accepts_single_action_wrapper() -> None: |
| 91 | request = { |
| 92 | "thoughts": ["Read the requested file."], |
| 93 | "actions": [ |
| 94 | { |
| 95 | "tool_name": "text_editor", |
| 96 | "tool_args": {"action": "read", "path": "README.md"}, |
| 97 | } |
| 98 | ], |
| 99 | } |
| 100 | |
| 101 | assert json_parse_dirty(str(request)) == request |
| 102 | assert normalize_tool_request(request) == ( |
| 103 | "text_editor", |
| 104 | {"action": "read", "path": "README.md"}, |
| 105 | ) |
| 106 | |
| 107 | |
| 108 | def test_normalize_tool_request_rejects_multiple_wrapped_actions() -> None: |
| 109 | with pytest.raises(ValueError, match="exactly one"): |
| 110 | normalize_tool_request( |
| 111 | { |
| 112 | "actions": [ |
| 113 | {"tool_name": "response", "tool_args": {"text": "first"}}, |
| 114 | {"tool_name": "response", "tool_args": {"text": "second"}}, |
| 115 | ] |
| 116 | } |
| 117 | ) |
| 118 | |
| 119 | |
| 120 | def test_extract_tool_request_requires_a_complete_tool_message() -> None: |
| 121 | request = '{"tool_name":"response","tool_args":{"text":"ok"}}' |
| 122 | |
| 123 | assert extract_tool_request(request) == { |
| 124 | "tool_name": "response", |
| 125 | "tool_args": {"text": "ok"}, |
| 126 | } |
| 127 | assert extract_tool_request('{"status":"ok"}') is None |
| 128 | assert extract_tool_request(f"Example: {request}") is None |
| 129 | assert extract_tool_request(f"{request} trailing text") is None |
| 130 | |
| 131 | |
| 132 | def test_is_misformatted_tool_request_requires_agent_tool_envelope() -> None: |
| 133 | request = '{"tool_name":"response","tool_args":{"text":"ok"}}' |
| 134 | concatenated = ( |
| 135 | '{"thoughts":[],"headline":"Inspecting","tool_name":"code_execution_tool",' |
| 136 | '"tool_args":{"code":"pwd"}}' |
| 137 | '{"thoughts":[],"headline":"Answering","tool_name":"response",' |
| 138 | '"tool_args":{"text":"done"}}' |
| 139 | ) |
| 140 | malformed = ( |
| 141 | '{"thoughts":["Plan the work", "Run the tools", ' |
| 142 | '"headline":"Save results", "tool_name":"parallel", ' |
| 143 | '"tool_args":{"tool_calls":[{"tool_name":"memory_save",' |
| 144 | '"tool_args":{"text":"ok"}}],"wait":true}}' |
| 145 | ) |
| 146 | |
| 147 | assert extract_tool_request(malformed) is None |
| 148 | assert is_misformatted_tool_request(malformed) is True |
| 149 | assert extract_tool_request(concatenated) is None |
| 150 | assert is_misformatted_tool_request(concatenated) is True |
| 151 | assert is_misformatted_tool_request(f"Intro\n```json\n{request}\n```") is True |
| 152 | assert is_misformatted_tool_request('{"status":"planning"}') is False |
| 153 | assert is_misformatted_tool_request(f"Example: {request}") is False |
| 154 | assert is_misformatted_tool_request( |
| 155 | malformed.replace('{"thoughts"', '{"status":"planning","thoughts"') |
| 156 | ) is False |
| 157 | |
| 158 | |
| 159 | def test_normalize_parallel_tool_calls_accepts_full_agent_reply_shape() -> None: |
| 160 | calls = parallel_tools.normalize_parallel_tool_calls( |
| 161 | [ |
| 162 | { |
| 163 | "thoughts": ["This is independent and ready to run."], |
| 164 | "headline": "Search Python release notes", |
| 165 | "tool_name": "search_engine", |
| 166 | "tool_args": {"query": "latest Python version changelog"}, |
| 167 | } |
| 168 | ] |
| 169 | ) |
| 170 | |
| 171 | assert calls[0].tool_name == "search_engine" |
| 172 | assert calls[0].tool_args == {"query": "latest Python version changelog"} |
| 173 | |
| 174 | |
| 175 | def test_parallel_prompt_encourages_mixed_independent_batches() -> None: |
| 176 | prompt = (PROJECT_ROOT / "prompts" / "agent.system.tool.parallel.md").read_text( |
| 177 | encoding="utf-8" |
| 178 | ) |
| 179 | |
| 180 | assert "same `tool_name` and `tool_args` shape as a top-level reply" in prompt |
| 181 | assert "planning fields like `thoughts` or `headline` are ignored" in prompt |
| 182 | assert "even when they use different tools" in prompt |
| 183 | assert "Do not split by tool type" in prompt |
| 184 | assert "Never include `document_query`" in prompt |
| 185 | assert "Call `response` only as a top-level tool" in prompt |