| 1 | from __future__ import annotations |
| 2 | |
| 3 | import asyncio |
| 4 | import importlib |
| 5 | import sys |
| 6 | import types |
| 7 | from dataclasses import dataclass |
| 8 | from pathlib import Path |
| 9 | |
| 10 | |
| 11 | @dataclass |
| 12 | class _FakeResponse: |
| 13 | message: str |
| 14 | break_loop: bool |
| 15 | additional: dict | None = None |
| 16 | |
| 17 | |
| 18 | class _FakeTool: |
| 19 | def __init__( |
| 20 | self, |
| 21 | agent, |
| 22 | name: str, |
| 23 | method: str | None, |
| 24 | args: dict | None, |
| 25 | message: str, |
| 26 | loop_data=None, |
| 27 | **kwargs, |
| 28 | ) -> None: |
| 29 | self.agent = agent |
| 30 | self.name = name |
| 31 | self.method = method |
| 32 | self.args = args or {} |
| 33 | self.message = message |
| 34 | self.loop_data = loop_data |
| 35 | |
| 36 | |
| 37 | class _FakeContext: |
| 38 | def __init__(self, data: dict | None = None) -> None: |
| 39 | self.id = "ctx" |
| 40 | self.data = data or {} |
| 41 | |
| 42 | def get_data(self, key, recursive=True): |
| 43 | return self.data.get(key) |
| 44 | |
| 45 | def set_data(self, key, value, recursive=True): |
| 46 | self.data[key] = value |
| 47 | |
| 48 | |
| 49 | class _FakeAgent: |
| 50 | def __init__(self) -> None: |
| 51 | self.data = {} |
| 52 | self.context = _FakeContext() |
| 53 | self.history = types.SimpleNamespace(output=lambda: []) |
| 54 | |
| 55 | def read_prompt(self, _name: str, **kwargs) -> str: |
| 56 | return f"deleted {kwargs.get('memory_count', 0)}" |
| 57 | |
| 58 | |
| 59 | @dataclass |
| 60 | class _FakeSkill: |
| 61 | name: str |
| 62 | description: str |
| 63 | path: Path |
| 64 | version: str = "" |
| 65 | tags: list[str] | None = None |
| 66 | |
| 67 | |
| 68 | def _normalize_loaded_skill_names(raw) -> list[str]: |
| 69 | if not isinstance(raw, list): |
| 70 | return [] |
| 71 | return [name for value in raw if (name := str(value or "").strip())] |
| 72 | |
| 73 | |
| 74 | def _get_loaded_skill_names(agent) -> list[str]: |
| 75 | names = _normalize_loaded_skill_names(agent.context.get_data("loaded_skills")) |
| 76 | if names: |
| 77 | return names |
| 78 | names = _normalize_loaded_skill_names(agent.data.get("loaded_skills")) |
| 79 | if names: |
| 80 | _set_loaded_skill_names(agent, names) |
| 81 | return names |
| 82 | |
| 83 | |
| 84 | def _set_loaded_skill_names(agent, names) -> list[str]: |
| 85 | names = _normalize_loaded_skill_names(names) |
| 86 | agent.context.set_data("loaded_skills", names or None) |
| 87 | return names |
| 88 | |
| 89 | |
| 90 | def _add_loaded_skill_name(agent, skill_name, *, limit=None) -> list[str]: |
| 91 | skill_name = str(skill_name or "").strip() |
| 92 | names = [name for name in _get_loaded_skill_names(agent) if name != skill_name] |
| 93 | if skill_name: |
| 94 | names.append(skill_name) |
| 95 | return _set_loaded_skill_names(agent, names[-(limit or 20):]) |
| 96 | |
| 97 | |
| 98 | def _skill_instruction_name(message) -> str: |
| 99 | match message: |
| 100 | case { |
| 101 | "content": { |
| 102 | "skill_instructions": { |
| 103 | "content_included": included, |
| 104 | "name": name, |
| 105 | } |
| 106 | } |
| 107 | } if included: |
| 108 | return str(name or "").strip() |
| 109 | return "" |
| 110 | |
| 111 | |
| 112 | def _install_tool_stub(monkeypatch) -> None: |
| 113 | tool_stub = types.ModuleType("helpers.tool") |
| 114 | tool_stub.Tool = _FakeTool |
| 115 | tool_stub.Response = _FakeResponse |
| 116 | monkeypatch.setitem(sys.modules, "helpers.tool", tool_stub) |
| 117 | |
| 118 | |
| 119 | def _load_skills_tool(monkeypatch, skill_root: Path): |
| 120 | _install_tool_stub(monkeypatch) |
| 121 | |
| 122 | skills_stub = types.ModuleType("helpers.skills") |
| 123 | skills_stub.AGENT_DATA_NAME_LOADED_SKILLS = "loaded_skills" |
| 124 | skills_stub.CONTEXT_DATA_NAME_LOADED_SKILLS = "loaded_skills" |
| 125 | skills_stub.MAX_ACTIVE_SKILLS = 20 |
| 126 | fake_skill = _FakeSkill( |
| 127 | name="browser-form-workflows", |
| 128 | description="Use for complex browser forms.", |
| 129 | path=skill_root, |
| 130 | tags=[], |
| 131 | ) |
| 132 | skills_stub.list_skills = lambda *args, **kwargs: [fake_skill] |
| 133 | skills_stub.search_skills = lambda *args, **kwargs: [fake_skill] |
| 134 | skills_stub.find_skill = lambda *args, **kwargs: fake_skill |
| 135 | skills_stub.load_skill_for_agent = ( |
| 136 | lambda *args, **kwargs: "Skill: browser-form-workflows\n\nInstructions:\nUse labels before typing." |
| 137 | ) |
| 138 | skills_stub.add_loaded_skill_name = _add_loaded_skill_name |
| 139 | skills_stub.get_loaded_skill_names = _get_loaded_skill_names |
| 140 | skills_stub.set_loaded_skill_names = _set_loaded_skill_names |
| 141 | skills_stub.skill_instruction_name = _skill_instruction_name |
| 142 | monkeypatch.setitem(sys.modules, "helpers.skills", skills_stub) |
| 143 | import helpers |
| 144 | monkeypatch.setattr(helpers, "skills", skills_stub, raising=False) |
| 145 | |
| 146 | print_style_stub = types.ModuleType("helpers.print_style") |
| 147 | print_style_stub.PrintStyle = lambda *args, **kwargs: types.SimpleNamespace( |
| 148 | print=lambda *a, **k: None |
| 149 | ) |
| 150 | monkeypatch.setitem(sys.modules, "helpers.print_style", print_style_stub) |
| 151 | |
| 152 | sys.modules.pop("tools.skills_tool", None) |
| 153 | return importlib.import_module("tools.skills_tool") |
| 154 | |
| 155 | |
| 156 | class _FakeExtension: |
| 157 | def __init__(self, agent=None): |
| 158 | self.agent = agent |
| 159 | |
| 160 | |
| 161 | class _FakeLoadedSkillAgent: |
| 162 | def __init__(self) -> None: |
| 163 | self.data = {} |
| 164 | self.context = _FakeContext({"loaded_skills": ["browser-form-workflows"]}) |
| 165 | self.added_tool_results = [] |
| 166 | |
| 167 | def hist_add_tool_result(self, tool_name: str, tool_result: str, **kwargs): |
| 168 | content = {"tool_name": tool_name, "tool_result": tool_result, **kwargs} |
| 169 | self.added_tool_results.append(content) |
| 170 | return types.SimpleNamespace( |
| 171 | output=lambda: [{"ai": False, "content": content}] |
| 172 | ) |
| 173 | |
| 174 | |
| 175 | def _load_loaded_skills_extension(monkeypatch, skill_root: Path): |
| 176 | extension_stub = types.ModuleType("helpers.extension") |
| 177 | extension_stub.Extension = _FakeExtension |
| 178 | monkeypatch.setitem(sys.modules, "helpers.extension", extension_stub) |
| 179 | |
| 180 | agent_stub = types.ModuleType("agent") |
| 181 | agent_stub.LoopData = lambda **kwargs: types.SimpleNamespace(**kwargs) |
| 182 | monkeypatch.setitem(sys.modules, "agent", agent_stub) |
| 183 | |
| 184 | skills_stub = types.ModuleType("helpers.skills") |
| 185 | fake_skill = _FakeSkill( |
| 186 | name="browser-form-workflows", |
| 187 | description="Use for complex browser forms.", |
| 188 | path=skill_root, |
| 189 | tags=[], |
| 190 | ) |
| 191 | skills_stub.find_skill = lambda *args, **kwargs: fake_skill |
| 192 | skills_stub.load_skill_for_agent = ( |
| 193 | lambda *args, **kwargs: "Skill: browser-form-workflows\n\nInstructions:\nUse labels before typing." |
| 194 | ) |
| 195 | skills_stub.get_loaded_skill_names = _get_loaded_skill_names |
| 196 | skills_stub.set_loaded_skill_names = _set_loaded_skill_names |
| 197 | skills_stub.skill_instruction_name = _skill_instruction_name |
| 198 | monkeypatch.setitem(sys.modules, "helpers.skills", skills_stub) |
| 199 | |
| 200 | tokens_stub = types.ModuleType("helpers.tokens") |
| 201 | tokens_stub.approximate_tokens = lambda text: len(str(text).split()) |
| 202 | monkeypatch.setitem(sys.modules, "helpers.tokens", tokens_stub) |
| 203 | |
| 204 | import helpers |
| 205 | |
| 206 | monkeypatch.setattr(helpers, "skills", skills_stub, raising=False) |
| 207 | monkeypatch.setattr(helpers, "tokens", tokens_stub, raising=False) |
| 208 | |
| 209 | skills_tool_stub = types.ModuleType("tools.skills_tool") |
| 210 | skills_tool_stub.DATA_NAME_LOADED_SKILLS = "loaded_skills" |
| 211 | monkeypatch.setitem(sys.modules, "tools.skills_tool", skills_tool_stub) |
| 212 | |
| 213 | module_name = "extensions.python.message_loop_prompts_after._65_include_loaded_skills" |
| 214 | sys.modules.pop(module_name, None) |
| 215 | return importlib.import_module(module_name) |
| 216 | |
| 217 | |
| 218 | def _load_relevant_skills_extension( |
| 219 | monkeypatch, queries: list[str], *, skills_tool_allowed: bool = True |
| 220 | ): |
| 221 | extension_stub = types.ModuleType("helpers.extension") |
| 222 | extension_stub.Extension = _FakeExtension |
| 223 | monkeypatch.setitem(sys.modules, "helpers.extension", extension_stub) |
| 224 | |
| 225 | agent_stub = types.ModuleType("agent") |
| 226 | agent_stub.LoopData = lambda **kwargs: types.SimpleNamespace(**kwargs) |
| 227 | monkeypatch.setitem(sys.modules, "agent", agent_stub) |
| 228 | |
| 229 | skills_stub = types.ModuleType("helpers.skills") |
| 230 | |
| 231 | def _search_skills(query, *args, **kwargs): |
| 232 | queries.append(query) |
| 233 | return [] |
| 234 | |
| 235 | skills_stub.search_skills = _search_skills |
| 236 | monkeypatch.setitem(sys.modules, "helpers.skills", skills_stub) |
| 237 | |
| 238 | tool_policy_stub = types.ModuleType("helpers.tool_policy") |
| 239 | tool_policy_stub.resolve_tool = lambda *args, **kwargs: types.SimpleNamespace( |
| 240 | allowed=skills_tool_allowed |
| 241 | ) |
| 242 | monkeypatch.setitem(sys.modules, "helpers.tool_policy", tool_policy_stub) |
| 243 | |
| 244 | import helpers |
| 245 | |
| 246 | monkeypatch.setattr(helpers, "skills", skills_stub, raising=False) |
| 247 | monkeypatch.setattr(helpers, "tool_policy", tool_policy_stub, raising=False) |
| 248 | |
| 249 | module_name = "extensions.python.message_loop_prompts_after._63_recall_relevant_skills" |
| 250 | sys.modules.pop(module_name, None) |
| 251 | return importlib.import_module(module_name) |
| 252 | |
| 253 | |
| 254 | def _load_computer_use_remote_tool(monkeypatch): |
| 255 | _install_tool_stub(monkeypatch) |
| 256 | |
| 257 | history_stub = types.ModuleType("helpers.history") |
| 258 | |
| 259 | class _RawMessage(dict): |
| 260 | def __init__(self, raw_content, preview): |
| 261 | super().__init__(raw_content=raw_content, preview=preview) |
| 262 | |
| 263 | history_stub.RawMessage = _RawMessage |
| 264 | monkeypatch.setitem(sys.modules, "helpers.history", history_stub) |
| 265 | |
| 266 | print_style_stub = types.ModuleType("helpers.print_style") |
| 267 | print_style_stub.PrintStyle = lambda *args, **kwargs: types.SimpleNamespace( |
| 268 | print=lambda *a, **k: None |
| 269 | ) |
| 270 | monkeypatch.setitem(sys.modules, "helpers.print_style", print_style_stub) |
| 271 | |
| 272 | ws_stub = types.ModuleType("helpers.ws") |
| 273 | ws_stub.NAMESPACE = "/test" |
| 274 | monkeypatch.setitem(sys.modules, "helpers.ws", ws_stub) |
| 275 | |
| 276 | ws_manager_stub = types.ModuleType("helpers.ws_manager") |
| 277 | ws_manager_stub.ConnectionNotFoundError = RuntimeError |
| 278 | ws_manager_stub.get_shared_ws_manager = lambda: types.SimpleNamespace( |
| 279 | emit_to=lambda *a, **k: None |
| 280 | ) |
| 281 | monkeypatch.setitem(sys.modules, "helpers.ws_manager", ws_manager_stub) |
| 282 | |
| 283 | ws_runtime_stub = types.ModuleType("plugins._a0_connector.helpers.ws_runtime") |
| 284 | ws_runtime_stub.clear_pending_computer_use_op = lambda *args, **kwargs: None |
| 285 | ws_runtime_stub.computer_use_metadata_for_sid = lambda *args, **kwargs: {} |
| 286 | ws_runtime_stub.select_computer_use_target_sid = lambda *args, **kwargs: "sid" |
| 287 | ws_runtime_stub.store_pending_computer_use_op = lambda *args, **kwargs: None |
| 288 | monkeypatch.setitem( |
| 289 | sys.modules, |
| 290 | "plugins._a0_connector.helpers.ws_runtime", |
| 291 | ws_runtime_stub, |
| 292 | ) |
| 293 | |
| 294 | sys.modules.pop("plugins._a0_connector.tools.computer_use_remote", None) |
| 295 | return importlib.import_module("plugins._a0_connector.tools.computer_use_remote") |
| 296 | |
| 297 | |
| 298 | def test_skills_tool_accepts_action_alias_for_search(monkeypatch, tmp_path: Path): |
| 299 | module = _load_skills_tool(monkeypatch, tmp_path) |
| 300 | tool = module.SkillsTool( |
| 301 | _FakeAgent(), |
| 302 | "skills_tool", |
| 303 | None, |
| 304 | {"action": "search", "query": "browser forms"}, |
| 305 | "", |
| 306 | None, |
| 307 | ) |
| 308 | |
| 309 | response = asyncio.run(tool.execute(**tool.args)) |
| 310 | |
| 311 | assert "browser-form-workflows" in response.message |
| 312 | |
| 313 | |
| 314 | def test_skills_tool_accepts_method_as_deprecated_action_alias( |
| 315 | monkeypatch, tmp_path: Path |
| 316 | ): |
| 317 | module = _load_skills_tool(monkeypatch, tmp_path) |
| 318 | tool = module.SkillsTool( |
| 319 | _FakeAgent(), |
| 320 | "skills_tool", |
| 321 | None, |
| 322 | {"method": "search", "query": "browser forms"}, |
| 323 | "", |
| 324 | None, |
| 325 | ) |
| 326 | |
| 327 | response = asyncio.run(tool.execute(**tool.args)) |
| 328 | |
| 329 | assert "browser-form-workflows" in response.message |
| 330 | assert tool.args["action"] == "search" |
| 331 | |
| 332 | |
| 333 | def test_skills_tool_defaults_missing_action_to_list(monkeypatch, tmp_path: Path): |
| 334 | module = _load_skills_tool(monkeypatch, tmp_path) |
| 335 | tool = module.SkillsTool( |
| 336 | _FakeAgent(), |
| 337 | "skills_tool", |
| 338 | None, |
| 339 | {}, |
| 340 | "", |
| 341 | None, |
| 342 | ) |
| 343 | |
| 344 | response = asyncio.run(tool.execute()) |
| 345 | |
| 346 | assert "Available skills" in response.message |
| 347 | assert "browser-form-workflows" in response.message |
| 348 | assert "slash commands" not in response.message |
| 349 | |
| 350 | |
| 351 | def test_skills_tool_load_appends_skill_instructions_as_tool_result( |
| 352 | monkeypatch, tmp_path: Path |
| 353 | ): |
| 354 | module = _load_skills_tool(monkeypatch, tmp_path) |
| 355 | agent = _FakeAgent() |
| 356 | tool = module.SkillsTool( |
| 357 | agent, |
| 358 | "skills_tool", |
| 359 | None, |
| 360 | {"action": "load", "skill_name": "browser-form-workflows"}, |
| 361 | "", |
| 362 | None, |
| 363 | ) |
| 364 | |
| 365 | response = asyncio.run(tool.execute(**tool.args)) |
| 366 | |
| 367 | assert "Skill: browser-form-workflows" in response.message |
| 368 | assert response.additional["skill_instructions"]["name"] == "browser-form-workflows" |
| 369 | assert response.additional["skill_instructions"]["content_included"] is True |
| 370 | assert agent.context.get_data("loaded_skills") == ["browser-form-workflows"] |
| 371 | |
| 372 | |
| 373 | def test_skills_tool_load_omits_duplicate_visible_skill( |
| 374 | monkeypatch, tmp_path: Path |
| 375 | ): |
| 376 | module = _load_skills_tool(monkeypatch, tmp_path) |
| 377 | agent = _FakeAgent() |
| 378 | tool = module.SkillsTool( |
| 379 | agent, |
| 380 | "skills_tool", |
| 381 | None, |
| 382 | {"action": "load", "skill_name": "browser-form-workflows"}, |
| 383 | "", |
| 384 | None, |
| 385 | ) |
| 386 | first = asyncio.run(tool.execute(**tool.args)) |
| 387 | loaded_message = { |
| 388 | "ai": False, |
| 389 | "content": {"skill_instructions": first.additional["skill_instructions"]}, |
| 390 | } |
| 391 | agent.history = types.SimpleNamespace(output=lambda: [loaded_message]) |
| 392 | |
| 393 | second = asyncio.run(tool.execute(**tool.args)) |
| 394 | |
| 395 | assert "already loaded in visible chat history" in second.message |
| 396 | assert "Instructions:\nUse labels before typing." not in second.message |
| 397 | assert second.additional["skill_instructions"]["content_included"] is False |
| 398 | assert second.additional["skill_instructions"]["already_loaded"] is True |
| 399 | assert agent.context.get_data("loaded_skills") == ["browser-form-workflows"] |
| 400 | |
| 401 | |
| 402 | def test_skills_tool_load_reloads_when_prior_skill_is_not_model_visible( |
| 403 | monkeypatch, tmp_path: Path |
| 404 | ): |
| 405 | module = _load_skills_tool(monkeypatch, tmp_path) |
| 406 | agent = _FakeAgent() |
| 407 | tool = module.SkillsTool( |
| 408 | agent, |
| 409 | "skills_tool", |
| 410 | None, |
| 411 | {"action": "load", "skill_name": "browser-form-workflows"}, |
| 412 | "", |
| 413 | None, |
| 414 | ) |
| 415 | first = asyncio.run(tool.execute(**tool.args)) |
| 416 | hidden_message = types.SimpleNamespace( |
| 417 | summary="", |
| 418 | content={"skill_instructions": first.additional["skill_instructions"]}, |
| 419 | ) |
| 420 | agent.history = types.SimpleNamespace( |
| 421 | all_messages=lambda: [hidden_message], |
| 422 | output=lambda: [ |
| 423 | { |
| 424 | "ai": False, |
| 425 | "content": "Earlier history was summarized and no skill body is visible.", |
| 426 | } |
| 427 | ], |
| 428 | ) |
| 429 | |
| 430 | second = asyncio.run(tool.execute(**tool.args)) |
| 431 | |
| 432 | assert "Skill: browser-form-workflows" in second.message |
| 433 | assert second.additional["skill_instructions"]["content_included"] is True |
| 434 | assert agent.context.get_data("loaded_skills") == ["browser-form-workflows"] |
| 435 | |
| 436 | |
| 437 | def test_loaded_skills_extension_reattaches_missing_body_after_compaction( |
| 438 | monkeypatch, tmp_path: Path |
| 439 | ): |
| 440 | module = _load_loaded_skills_extension(monkeypatch, tmp_path) |
| 441 | agent = _FakeLoadedSkillAgent() |
| 442 | loop_data = types.SimpleNamespace( |
| 443 | protocol_persistent={}, |
| 444 | extras_persistent={}, |
| 445 | history_output=[ |
| 446 | { |
| 447 | "ai": False, |
| 448 | "content": "Earlier history was summarized and no skill body is visible.", |
| 449 | } |
| 450 | ], |
| 451 | ) |
| 452 | |
| 453 | asyncio.run(module.IncludeLoadedSkills(agent).execute(loop_data)) |
| 454 | |
| 455 | assert len(agent.added_tool_results) == 1 |
| 456 | added = agent.added_tool_results[0] |
| 457 | assert added["tool_name"] == "skills_tool" |
| 458 | assert "Skill: browser-form-workflows" in added["tool_result"] |
| 459 | assert added["skill_instructions"] == { |
| 460 | "name": "browser-form-workflows", |
| 461 | "path": str(tmp_path), |
| 462 | "source": "skills_tool:reattach", |
| 463 | "content_included": True, |
| 464 | } |
| 465 | assert loop_data.history_output[-1]["content"] == added |
| 466 | |
| 467 | |
| 468 | def test_loaded_skills_extension_does_not_reattach_visible_skill( |
| 469 | monkeypatch, tmp_path: Path |
| 470 | ): |
| 471 | module = _load_loaded_skills_extension(monkeypatch, tmp_path) |
| 472 | agent = _FakeLoadedSkillAgent() |
| 473 | loop_data = types.SimpleNamespace( |
| 474 | protocol_persistent={}, |
| 475 | extras_persistent={}, |
| 476 | history_output=[ |
| 477 | { |
| 478 | "ai": False, |
| 479 | "content": { |
| 480 | "skill_instructions": { |
| 481 | "name": "browser-form-workflows", |
| 482 | "content_included": True, |
| 483 | } |
| 484 | }, |
| 485 | } |
| 486 | ], |
| 487 | ) |
| 488 | |
| 489 | asyncio.run(module.IncludeLoadedSkills(agent).execute(loop_data)) |
| 490 | |
| 491 | assert agent.added_tool_results == [] |
| 492 | |
| 493 | |
| 494 | def test_loaded_skills_extension_keeps_reattachments_under_budget( |
| 495 | monkeypatch, tmp_path: Path |
| 496 | ): |
| 497 | module = _load_loaded_skills_extension(monkeypatch, tmp_path) |
| 498 | monkeypatch.setattr(module, "SKILL_REATTACHMENT_TOKEN_BUDGET", 1) |
| 499 | agent = _FakeLoadedSkillAgent() |
| 500 | loop_data = types.SimpleNamespace( |
| 501 | protocol_persistent={}, |
| 502 | extras_persistent={}, |
| 503 | history_output=[], |
| 504 | ) |
| 505 | |
| 506 | asyncio.run(module.IncludeLoadedSkills(agent).execute(loop_data)) |
| 507 | |
| 508 | assert agent.added_tool_results == [] |
| 509 | |
| 510 | |
| 511 | def test_relevant_skill_recall_uses_raw_user_message(monkeypatch): |
| 512 | queries: list[str] = [] |
| 513 | module = _load_relevant_skills_extension(monkeypatch, queries) |
| 514 | agent = types.SimpleNamespace() |
| 515 | loop_data = types.SimpleNamespace( |
| 516 | iteration=0, |
| 517 | user_message=types.SimpleNamespace( |
| 518 | content={ |
| 519 | "system_message": [], |
| 520 | "user_message": "Open a browser and take a screenshot.", |
| 521 | "attachments": [], |
| 522 | }, |
| 523 | output_text=lambda: 'user: {"user_message": "wrapped"}', |
| 524 | ), |
| 525 | extras_temporary={}, |
| 526 | ) |
| 527 | |
| 528 | asyncio.run(module.RecallRelevantSkills(agent).execute(loop_data=loop_data)) |
| 529 | |
| 530 | assert queries == ["Open a browser and take a screenshot."] |
| 531 | |
| 532 | |
| 533 | def test_relevant_skill_recall_skips_blocked_skills_tool(monkeypatch): |
| 534 | queries: list[str] = [] |
| 535 | module = _load_relevant_skills_extension( |
| 536 | monkeypatch, queries, skills_tool_allowed=False |
| 537 | ) |
| 538 | loop_data = types.SimpleNamespace( |
| 539 | iteration=0, |
| 540 | user_message=types.SimpleNamespace( |
| 541 | content="Open a browser and take a screenshot.", |
| 542 | output_text=lambda: "Open a browser and take a screenshot.", |
| 543 | ), |
| 544 | extras_temporary={}, |
| 545 | ) |
| 546 | |
| 547 | asyncio.run( |
| 548 | module.RecallRelevantSkills(types.SimpleNamespace()).execute( |
| 549 | loop_data=loop_data |
| 550 | ) |
| 551 | ) |
| 552 | |
| 553 | assert queries == [] |
| 554 | |
| 555 | |
| 556 | def test_skills_tool_read_file_action_reads_inside_skill_dir( |
| 557 | monkeypatch, tmp_path: Path |
| 558 | ): |
| 559 | skill_root = tmp_path / "browser-form-workflows" |
| 560 | skill_root.mkdir() |
| 561 | (skill_root / "notes.md").write_text("Use labels before typing.\n", encoding="utf-8") |
| 562 | module = _load_skills_tool(monkeypatch, skill_root) |
| 563 | tool = module.SkillsTool( |
| 564 | _FakeAgent(), |
| 565 | "skills_tool", |
| 566 | None, |
| 567 | { |
| 568 | "action": "read_file", |
| 569 | "skill_name": "browser-form-workflows", |
| 570 | "file_path": "notes.md", |
| 571 | }, |
| 572 | "", |
| 573 | None, |
| 574 | ) |
| 575 | |
| 576 | response = asyncio.run(tool.execute(**tool.args)) |
| 577 | |
| 578 | assert "Skill file: browser-form-workflows/notes.md" in response.message |
| 579 | assert "Use labels before typing." in response.message |
| 580 | |
| 581 | |
| 582 | def test_memory_forget_tool_imports_plugin_memory_load(monkeypatch): |
| 583 | _install_tool_stub(monkeypatch) |
| 584 | monkeypatch.syspath_prepend(str(Path.cwd())) |
| 585 | |
| 586 | class FakeDb: |
| 587 | def __init__(self) -> None: |
| 588 | self.calls = [] |
| 589 | |
| 590 | async def delete_documents_by_query(self, **kwargs): |
| 591 | self.calls.append(kwargs) |
| 592 | return ["memory-1"] |
| 593 | |
| 594 | fake_db = FakeDb() |
| 595 | |
| 596 | async def get_memory(_agent): |
| 597 | return fake_db |
| 598 | |
| 599 | memory_stub = types.ModuleType("plugins._memory.helpers.memory") |
| 600 | memory_stub.Memory = types.SimpleNamespace(get=get_memory) |
| 601 | monkeypatch.setitem(sys.modules, "plugins._memory.helpers.memory", memory_stub) |
| 602 | |
| 603 | sys.modules.pop("plugins._memory.tools.memory_load", None) |
| 604 | sys.modules.pop("plugins._memory.tools.memory_forget", None) |
| 605 | module = importlib.import_module("plugins._memory.tools.memory_forget") |
| 606 | tool = module.MemoryForget( |
| 607 | _FakeAgent(), |
| 608 | "memory_forget", |
| 609 | None, |
| 610 | { |
| 611 | "query": "codex memory forget token", |
| 612 | "threshold": 0.99, |
| 613 | "filter": "area=='codex_sweep'", |
| 614 | }, |
| 615 | "", |
| 616 | None, |
| 617 | ) |
| 618 | |
| 619 | response = asyncio.run(tool.execute(**tool.args)) |
| 620 | |
| 621 | assert response.message == "deleted 1" |
| 622 | assert fake_db.calls == [ |
| 623 | { |
| 624 | "query": "codex memory forget token", |
| 625 | "threshold": 0.99, |
| 626 | "filter": "area=='codex_sweep'", |
| 627 | "include_exact": True, |
| 628 | "cascade": True, |
| 629 | } |
| 630 | ] |
| 631 | |
| 632 | |
| 633 | def test_memory_load_coerces_numeric_string_args(monkeypatch): |
| 634 | _install_tool_stub(monkeypatch) |
| 635 | monkeypatch.syspath_prepend(str(Path.cwd())) |
| 636 | |
| 637 | class FakeDb: |
| 638 | def __init__(self) -> None: |
| 639 | self.calls = [] |
| 640 | |
| 641 | async def search_similarity_threshold(self, **kwargs): |
| 642 | self.calls.append(kwargs) |
| 643 | return [] |
| 644 | |
| 645 | fake_db = FakeDb() |
| 646 | |
| 647 | async def get_memory(_agent): |
| 648 | return fake_db |
| 649 | |
| 650 | memory_stub = types.ModuleType("plugins._memory.helpers.memory") |
| 651 | memory_stub.Memory = types.SimpleNamespace(get=get_memory) |
| 652 | monkeypatch.setitem(sys.modules, "plugins._memory.helpers.memory", memory_stub) |
| 653 | |
| 654 | sys.modules.pop("plugins._memory.tools.memory_load", None) |
| 655 | module = importlib.import_module("plugins._memory.tools.memory_load") |
| 656 | tool = module.MemoryLoad( |
| 657 | _FakeAgent(), |
| 658 | "memory_load", |
| 659 | None, |
| 660 | { |
| 661 | "query": "smoke test project context", |
| 662 | "threshold": "0.7", |
| 663 | "limit": "3", |
| 664 | }, |
| 665 | "", |
| 666 | None, |
| 667 | ) |
| 668 | |
| 669 | asyncio.run(tool.execute(**tool.args)) |
| 670 | |
| 671 | assert fake_db.calls == [ |
| 672 | { |
| 673 | "query": "smoke test project context", |
| 674 | "threshold": 0.7, |
| 675 | "limit": 3, |
| 676 | "filter": "", |
| 677 | } |
| 678 | ] |
| 679 | |
| 680 | |
| 681 | def test_behaviour_adjustment_normalizes_duplicate_rules(monkeypatch): |
| 682 | _install_tool_stub(monkeypatch) |
| 683 | monkeypatch.syspath_prepend(str(Path.cwd())) |
| 684 | |
| 685 | agent_stub = types.ModuleType("agent") |
| 686 | agent_stub.Agent = object |
| 687 | monkeypatch.setitem(sys.modules, "agent", agent_stub) |
| 688 | |
| 689 | log_stub = types.ModuleType("helpers.log") |
| 690 | log_stub.LogItem = object |
| 691 | monkeypatch.setitem(sys.modules, "helpers.log", log_stub) |
| 692 | |
| 693 | memory_stub = types.ModuleType("plugins._memory.helpers.memory") |
| 694 | memory_stub.get_memory_subdir_abs = lambda agent: "/tmp" |
| 695 | monkeypatch.setitem(sys.modules, "plugins._memory.helpers.memory", memory_stub) |
| 696 | |
| 697 | sys.modules.pop("plugins._memory.tools.behaviour_adjustment", None) |
| 698 | module = importlib.import_module("plugins._memory.tools.behaviour_adjustment") |
| 699 | |
| 700 | rules = module.normalize_ruleset( |
| 701 | "## Behavioral rules\n" |
| 702 | "* Favor Linux commands.\n" |
| 703 | "* Token rule.## Behavioral rules\n" |
| 704 | "* Favor Linux commands.\n" |
| 705 | "* Token rule." |
| 706 | ) |
| 707 | |
| 708 | assert rules == "## Behavioral rules\n* Favor Linux commands.\n* Token rule.\n" |
| 709 | |
| 710 | |
| 711 | def test_behaviour_prompts_preserve_exact_rules_and_avoid_promptinclude(): |
| 712 | behaviour_prompt_path = Path( |
| 713 | "plugins/_memory/prompts/agent.system.tool.behaviour.md" |
| 714 | ) |
| 715 | behaviour_prompt = behaviour_prompt_path.read_text(encoding="utf-8") |
| 716 | merge_prompt = Path("prompts/behaviour.merge.sys.md").read_text( |
| 717 | encoding="utf-8" |
| 718 | ) |
| 719 | promptinclude_prompt = Path( |
| 720 | "plugins/_promptinclude/prompts/agent.system.promptinclude.md" |
| 721 | ).read_text(encoding="utf-8") |
| 722 | |
| 723 | assert "exact-response rules" in behaviour_prompt |
| 724 | assert "preserve it verbatim" in behaviour_prompt |
| 725 | assert not Path("prompts/agent.system.tool.behaviour.md").exists() |
| 726 | assert "respond exactly with a phrase" in merge_prompt |
| 727 | assert "use behaviour_adjustment, not promptinclude files" in promptinclude_prompt |
| 728 | |
| 729 | |
| 730 | def _load_a2a_chat_tool(monkeypatch): |
| 731 | _install_tool_stub(monkeypatch) |
| 732 | sys.modules.pop("tools.a2a_chat", None) |
| 733 | return importlib.import_module("tools.a2a_chat") |
| 734 | |
| 735 | |
| 736 | def test_a2a_extracts_latest_assistant_text_from_history(monkeypatch): |
| 737 | module = _load_a2a_chat_tool(monkeypatch) |
| 738 | |
| 739 | final = { |
| 740 | "result": { |
| 741 | "history": [ |
| 742 | { |
| 743 | "role": "user", |
| 744 | "parts": [{"kind": "text", "text": "what is 2+2?"}], |
| 745 | }, |
| 746 | { |
| 747 | "role": "assistant", |
| 748 | "parts": [{"kind": "text", "text": "4"}], |
| 749 | }, |
| 750 | ] |
| 751 | } |
| 752 | } |
| 753 | |
| 754 | assert module._extract_latest_assistant_text(final) == "4" |
| 755 | |
| 756 | |
| 757 | def test_a2a_extracts_status_or_artifact_text_when_history_is_empty(monkeypatch): |
| 758 | module = _load_a2a_chat_tool(monkeypatch) |
| 759 | |
| 760 | status_final = { |
| 761 | "result": { |
| 762 | "status": { |
| 763 | "message": { |
| 764 | "parts": [{"kind": "text", "text": "status answer"}] |
| 765 | } |
| 766 | } |
| 767 | } |
| 768 | } |
| 769 | artifact_final = { |
| 770 | "result": { |
| 771 | "artifacts": [ |
| 772 | {"parts": [{"kind": "text", "text": "artifact answer"}]} |
| 773 | ] |
| 774 | } |
| 775 | } |
| 776 | |
| 777 | assert module._extract_latest_assistant_text(status_final) == "status answer" |
| 778 | assert module._extract_latest_assistant_text(artifact_final) == "artifact answer" |
| 779 | |
| 780 | |
| 781 | def test_a2a_session_key_normalizes_explicit_a2a_path(monkeypatch): |
| 782 | module = _load_a2a_chat_tool(monkeypatch) |
| 783 | |
| 784 | assert module._session_key("http://localhost:32080/a2a") == "http://localhost:32080" |
| 785 | assert module._session_key("http://localhost:32080") == "http://localhost:32080" |
| 786 | |
| 787 | |
| 788 | def test_a2a_empty_response_message_is_explicit_failure(monkeypatch): |
| 789 | module = _load_a2a_chat_tool(monkeypatch) |
| 790 | |
| 791 | assert module._extract_latest_assistant_text({"result": {"history": []}}) == "" |
| 792 | assert "failed" in module.A2A_EMPTY_RESPONSE_ERROR |
| 793 | assert "not success" in module.A2A_EMPTY_RESPONSE_ERROR |
| 794 | |
| 795 | |
| 796 | def test_notify_user_prompt_documents_numeric_priority_values(): |
| 797 | prompt = Path("prompts/agent.system.tool.notify_user.md").read_text( |
| 798 | encoding="utf-8" |
| 799 | ) |
| 800 | |
| 801 | assert "priority values: `20` high urgency, `10` normal urgency" in prompt |
| 802 | |
| 803 | |
| 804 | def test_tool_prompts_prevent_top_level_multi_tool(): |
| 805 | tools_prompt = Path("prompts/agent.system.tools.md").read_text(encoding="utf-8") |
| 806 | communication_prompt = Path("prompts/agent.system.main.communication.md").read_text( |
| 807 | encoding="utf-8" |
| 808 | ) |
| 809 | browser_prompt = Path("plugins/_browser/prompts/agent.system.tool.browser.md").read_text( |
| 810 | encoding="utf-8" |
| 811 | ) |
| 812 | |
| 813 | assert "Do not invent top-level `multi` or generic batch tools" in tools_prompt |
| 814 | assert "listed wrapper for independent concurrent calls is `parallel`" in tools_prompt |
| 815 | assert "never an action name such as `read`, `write`, `terminal`, or `multi`" in communication_prompt |
| 816 | assert "independent operations concurrently" in communication_prompt |
| 817 | assert 'Never use `tool_name: "multi"`' in browser_prompt |
| 818 | |
| 819 | |
| 820 | def test_local_model_tool_use_guide_stays_prompt_profile_plugin_only(): |
| 821 | guide = Path("docs/guides/local-model-tool-use.md").read_text(encoding="utf-8") |
| 822 | |
| 823 | assert "Tiny Local" in guide |
| 824 | assert "agents/tiny-local/" in guide |
| 825 | assert "*.promptinclude.md" in guide |
| 826 | assert "Do not change `agent.py`" in guide |
| 827 | assert "Do not change `helpers/extract_tools.py`" in guide |
| 828 | assert "Use exactly these top-level fields: `tool_name` and `tool_args`." in guide |
| 829 | assert '{"tool_name":"response","tool_args":{"text":"Done."}}' in guide |
| 830 | assert "If the user says \"proceed\", \"continue\", \"go ahead\", or similar" in guide |
| 831 | assert "call the next appropriate tool instead of replying with a promise or status update" in guide |
| 832 | assert "Do not create parser repair code for this workflow." in guide |
| 833 | |
| 834 | |
| 835 | def _load_scheduler_tool(monkeypatch): |
| 836 | _install_tool_stub(monkeypatch) |
| 837 | |
| 838 | scheduler_stub = types.ModuleType("helpers.task_scheduler") |
| 839 | scheduler_stub.TaskScheduler = object |
| 840 | scheduler_stub.ScheduledTask = type("ScheduledTask", (), {}) |
| 841 | scheduler_stub.AdHocTask = type("AdHocTask", (), {}) |
| 842 | scheduler_stub.PlannedTask = type("PlannedTask", (), {}) |
| 843 | scheduler_stub.serialize_task = lambda task: {} |
| 844 | scheduler_stub.parse_datetime = lambda value: None |
| 845 | scheduler_stub.parse_task_plan = lambda value: None |
| 846 | scheduler_stub.serialize_datetime = lambda value: value |
| 847 | scheduler_stub.TaskState = types.SimpleNamespace( |
| 848 | IDLE="idle", |
| 849 | RUNNING="running", |
| 850 | ) |
| 851 | scheduler_stub.TaskSchedule = type("TaskSchedule", (), {}) |
| 852 | scheduler_stub.TaskPlan = type("TaskPlan", (), {}) |
| 853 | monkeypatch.setitem(sys.modules, "helpers.task_scheduler", scheduler_stub) |
| 854 | |
| 855 | agent_stub = types.ModuleType("agent") |
| 856 | agent_stub.AgentContext = types.SimpleNamespace( |
| 857 | get=lambda *args, **kwargs: None, |
| 858 | remove=lambda *args, **kwargs: None, |
| 859 | ) |
| 860 | monkeypatch.setitem(sys.modules, "agent", agent_stub) |
| 861 | |
| 862 | persist_chat_stub = types.ModuleType("helpers.persist_chat") |
| 863 | persist_chat_stub.remove_chat = lambda *args, **kwargs: None |
| 864 | monkeypatch.setitem(sys.modules, "helpers.persist_chat", persist_chat_stub) |
| 865 | |
| 866 | projects_stub = types.ModuleType("helpers.projects") |
| 867 | projects_stub.get_context_project_name = lambda context: "" |
| 868 | projects_stub.load_basic_project_data = lambda project: {} |
| 869 | monkeypatch.setitem(sys.modules, "helpers.projects", projects_stub) |
| 870 | |
| 871 | sys.modules.pop("tools.scheduler", None) |
| 872 | return importlib.import_module("tools.scheduler") |
| 873 | |
| 874 | |
| 875 | def test_scheduler_accepts_action_alias(monkeypatch): |
| 876 | module = _load_scheduler_tool(monkeypatch) |
| 877 | tool = module.SchedulerTool( |
| 878 | _FakeAgent(), |
| 879 | "scheduler", |
| 880 | None, |
| 881 | {"action": "list_tasks"}, |
| 882 | "", |
| 883 | None, |
| 884 | ) |
| 885 | |
| 886 | async def list_tasks(**kwargs): |
| 887 | return module.Response("listed", False) |
| 888 | |
| 889 | tool.list_tasks = list_tasks |
| 890 | |
| 891 | response = asyncio.run(tool.execute(**tool.args)) |
| 892 | |
| 893 | assert response.message == "listed" |
| 894 | |
| 895 | |
| 896 | def test_scheduler_requires_action_field(monkeypatch): |
| 897 | module = _load_scheduler_tool(monkeypatch) |
| 898 | tool = module.SchedulerTool( |
| 899 | _FakeAgent(), |
| 900 | "scheduler", |
| 901 | "list_tasks", |
| 902 | {}, |
| 903 | "", |
| 904 | None, |
| 905 | ) |
| 906 | |
| 907 | response = asyncio.run(tool.execute(**tool.args)) |
| 908 | |
| 909 | assert "Unknown scheduler action" in response.message |
| 910 | |
| 911 | |
| 912 | def test_scheduler_create_defaults_to_dedicated_context(monkeypatch): |
| 913 | module = _load_scheduler_tool(monkeypatch) |
| 914 | |
| 915 | class FakeTaskSchedule: |
| 916 | def __init__(self, **kwargs): |
| 917 | self.__dict__.update(kwargs) |
| 918 | |
| 919 | def to_crontab(self): |
| 920 | return f"{self.minute} {self.hour} {self.day} {self.month} {self.weekday}" |
| 921 | |
| 922 | class FakeScheduledTask: |
| 923 | @classmethod |
| 924 | def create(cls, **kwargs): |
| 925 | task = cls() |
| 926 | task.uuid = "task-1" |
| 927 | task.context_id = kwargs.get("context_id") |
| 928 | task.schedule = kwargs.get("schedule") |
| 929 | return task |
| 930 | |
| 931 | class FakeScheduler: |
| 932 | def __init__(self): |
| 933 | self.added = None |
| 934 | |
| 935 | async def add_task(self, task): |
| 936 | self.added = task |
| 937 | |
| 938 | fake_scheduler = FakeScheduler() |
| 939 | module.TaskSchedule = FakeTaskSchedule |
| 940 | module.ScheduledTask = FakeScheduledTask |
| 941 | module.TaskScheduler = types.SimpleNamespace(get=lambda: fake_scheduler) |
| 942 | tool = module.SchedulerTool( |
| 943 | _FakeAgent(), |
| 944 | "scheduler", |
| 945 | None, |
| 946 | { |
| 947 | "action": "create_scheduled_task", |
| 948 | "name": "check stuff", |
| 949 | "prompt": "tell me if anything changed", |
| 950 | "schedule": {"minute": "0", "hour": "9", "day": "*", "month": "*", "weekday": "*"}, |
| 951 | }, |
| 952 | "", |
| 953 | None, |
| 954 | ) |
| 955 | |
| 956 | response = asyncio.run(tool.execute(**tool.args)) |
| 957 | |
| 958 | assert "created" in response.message |
| 959 | assert fake_scheduler.added.context_id is None |
| 960 | |
| 961 | |
| 962 | def test_scheduler_local_timezone_alias_uses_current_user_timezone(monkeypatch): |
| 963 | module = _load_scheduler_tool(monkeypatch) |
| 964 | |
| 965 | class FakeTaskSchedule: |
| 966 | def __init__(self, **kwargs): |
| 967 | self.__dict__.update(kwargs) |
| 968 | |
| 969 | module.TaskSchedule = FakeTaskSchedule |
| 970 | module.Localization = types.SimpleNamespace( |
| 971 | get=lambda: types.SimpleNamespace(get_timezone=lambda: "Europe/Rome") |
| 972 | ) |
| 973 | |
| 974 | assert module._schedule_timezone({"schedule": {"timezone": "local"}}) == "Europe/Rome" |
| 975 | schedule = module._task_schedule_from_input( |
| 976 | {"minute": "30", "hour": "9", "day": "*", "month": "*", "weekday": "*", "timezone": "current"} |
| 977 | ) |
| 978 | |
| 979 | assert schedule.timezone == "Europe/Rome" |
| 980 | |
| 981 | |
| 982 | def test_scheduler_invalid_timezone_returns_repairable_message(monkeypatch): |
| 983 | module = _load_scheduler_tool(monkeypatch) |
| 984 | tool = module.SchedulerTool( |
| 985 | _FakeAgent(), |
| 986 | "scheduler", |
| 987 | None, |
| 988 | { |
| 989 | "action": "create_scheduled_task", |
| 990 | "name": "bad timezone", |
| 991 | "prompt": "tell me something", |
| 992 | "schedule": { |
| 993 | "minute": "0", |
| 994 | "hour": "9", |
| 995 | "day": "*", |
| 996 | "month": "*", |
| 997 | "weekday": "*", |
| 998 | "timezone": "Mars/Base", |
| 999 | }, |
| 1000 | }, |
| 1001 | "", |
| 1002 | None, |
| 1003 | ) |
| 1004 | |
| 1005 | response = asyncio.run(tool.execute(**tool.args)) |
| 1006 | |
| 1007 | assert "Invalid timezone: Mars/Base" in response.message |
| 1008 | |
| 1009 | |
| 1010 | def test_scheduler_prompt_includes_update_timezone_and_dedicated_context(): |
| 1011 | project_root = Path(__file__).resolve().parents[1] |
| 1012 | text = ( |
| 1013 | project_root / "prompts/agent.system.tool.scheduler.md" |
| 1014 | ).read_text(encoding="utf-8") |
| 1015 | |
| 1016 | assert "update_task" in text |
| 1017 | assert "timezone" in text |
| 1018 | assert "IANA" in text |
| 1019 | assert "dedicated context" in text |
| 1020 | |
| 1021 | |
| 1022 | def test_skills_prompt_renders_catalog_placeholder(): |
| 1023 | project_root = Path(__file__).resolve().parents[1] |
| 1024 | text = (project_root / "prompts/agent.system.skills.md").read_text( |
| 1025 | encoding="utf-8" |
| 1026 | ) |
| 1027 | |
| 1028 | assert "{{skills}}" in text |
| 1029 | |
| 1030 | |
| 1031 | def test_corrected_tool_prompts_only_teach_action_contract(): |
| 1032 | project_root = Path(__file__).resolve().parents[1] |
| 1033 | prompt_paths = [ |
| 1034 | project_root / "plugins/_text_editor/prompts/agent.system.tool.text_editor.md", |
| 1035 | project_root / "prompts/agent.system.tool.skills.md", |
| 1036 | project_root / "prompts/agent.system.tool.scheduler.md", |
| 1037 | project_root / "plugins/_a0_connector/prompts/agent.system.tool.text_editor_remote.md", |
| 1038 | project_root / "plugins/_office/prompts/agent.system.tool.office_artifact.md", |
| 1039 | project_root / "plugins/_office/skills/office-artifacts/SKILL.md", |
| 1040 | project_root / "plugins/_office/skills/markdown-documents/SKILL.md", |
| 1041 | project_root / "plugins/_office/skills/writer-documents/SKILL.md", |
| 1042 | project_root / "plugins/_office/skills/calc-spreadsheets/SKILL.md", |
| 1043 | project_root / "plugins/_office/skills/impress-presentations/SKILL.md", |
| 1044 | ] |
| 1045 | forbidden = ( |
| 1046 | "text_editor:", |
| 1047 | "skills_tool:", |
| 1048 | "scheduler:", |
| 1049 | "office_artifact:", |
| 1050 | "`method`", |
| 1051 | "`op`", |
| 1052 | "`operation`", |
| 1053 | "alias", |
| 1054 | ) |
| 1055 | |
| 1056 | for path in prompt_paths: |
| 1057 | text = path.read_text(encoding="utf-8") |
| 1058 | assert "action" in text |
| 1059 | for token in forbidden: |
| 1060 | assert token not in text |
| 1061 | if "document" in path.name or "_office/skills" in str(path): |
| 1062 | assert "faux UI action labels" in text |
| 1063 | assert "Open document" in text |
| 1064 | assert "Download file" in text |
| 1065 | assert "Canvas was not opened automatically" not in text |
| 1066 | assert "Open Document, or Desktop edit actions" not in text |
| 1067 | |
| 1068 | |
| 1069 | def test_computer_use_remote_is_runtime_checked_standard_tool(): |
| 1070 | project_root = Path(__file__).resolve().parents[1] |
| 1071 | standard_prompt_path = ( |
| 1072 | project_root |
| 1073 | / "plugins/_a0_connector/prompts/agent.system.tool.computer_use_remote.md" |
| 1074 | ) |
| 1075 | standard_prompt_text = standard_prompt_path.read_text(encoding="utf-8") |
| 1076 | skill_text = ( |
| 1077 | project_root |
| 1078 | / "plugins/_a0_connector/skills/host-computer-use/SKILL.md" |
| 1079 | ).read_text(encoding="utf-8") |
| 1080 | macos_skill_text = ( |
| 1081 | project_root |
| 1082 | / "plugins/_a0_connector/skills/host-computer-use-macos/SKILL.md" |
| 1083 | ).read_text(encoding="utf-8") |
| 1084 | windows_skill_text = ( |
| 1085 | project_root |
| 1086 | / "plugins/_a0_connector/skills/host-computer-use-windows/SKILL.md" |
| 1087 | ).read_text(encoding="utf-8") |
| 1088 | |
| 1089 | assert standard_prompt_path.exists() |
| 1090 | assert not ( |
| 1091 | project_root |
| 1092 | / "plugins/_a0_connector/prompts/agent.system.runtime_tool.computer_use_remote.md" |
| 1093 | ).exists() |
| 1094 | assert '"tool_name": "computer_use_remote"' in standard_prompt_text |
| 1095 | assert "not scoped to a single chat context" in standard_prompt_text |
| 1096 | assert "checked when the tool runs" in standard_prompt_text |
| 1097 | assert "visual verification is unavailable" in standard_prompt_text |
| 1098 | assert "host-computer-use-macos" in standard_prompt_text |
| 1099 | assert "host-computer-use-windows" in standard_prompt_text |
| 1100 | assert "ax_snapshot" not in standard_prompt_text |
| 1101 | assert "ax_action" not in standard_prompt_text |
| 1102 | assert "uia_snapshot" not in standard_prompt_text |
| 1103 | assert "uia_action" not in standard_prompt_text |
| 1104 | assert '"tool_name": "computer_use_remote"' in skill_text |
| 1105 | assert "ax_snapshot" not in skill_text |
| 1106 | assert "ax_action" not in skill_text |
| 1107 | assert "uia_snapshot" not in skill_text |
| 1108 | assert "uia_action" not in skill_text |
| 1109 | assert '"tool_name": "computer_use_remote"' in macos_skill_text |
| 1110 | assert "ax_snapshot" in macos_skill_text |
| 1111 | assert "ax_action" in macos_skill_text |
| 1112 | assert '"tool_name": "computer_use_remote"' in windows_skill_text |
| 1113 | assert "uia_snapshot" in windows_skill_text |
| 1114 | assert "uia_action" in windows_skill_text |
| 1115 | assert "focus_window" in windows_skill_text |
| 1116 | assert "If a node offers `invoke`, use `invoke`, not `click`" in windows_skill_text |
| 1117 | assert "Backend-specific macOS guidance" in macos_skill_text |
| 1118 | assert "Backend-specific Windows guidance" in windows_skill_text |
| 1119 | assert "Beta desktop control" in skill_text |
| 1120 | |
| 1121 | |
| 1122 | def test_computer_use_remote_start_session_reports_backend_features_and_macos_skill(monkeypatch): |
| 1123 | module = _load_computer_use_remote_tool(monkeypatch) |
| 1124 | tool = object.__new__(module.ComputerUseRemote) |
| 1125 | |
| 1126 | message = tool._extract_result( |
| 1127 | "start_session", |
| 1128 | { |
| 1129 | "ok": True, |
| 1130 | "result": { |
| 1131 | "session_id": "s1", |
| 1132 | "width": 1920, |
| 1133 | "height": 1080, |
| 1134 | "backend_id": "macos", |
| 1135 | "backend_family": "macos", |
| 1136 | "features": [ |
| 1137 | "accessibility-tree-snapshot", |
| 1138 | "accessibility-structural-targeting", |
| 1139 | ], |
| 1140 | }, |
| 1141 | }, |
| 1142 | ) |
| 1143 | |
| 1144 | assert "session_id=s1" in message |
| 1145 | assert "backend=macos/macos" in message |
| 1146 | assert "features=accessibility-tree-snapshot, accessibility-structural-targeting" in message |
| 1147 | assert "host-computer-use-macos" in message |
| 1148 | |
| 1149 | |
| 1150 | def test_computer_use_remote_start_session_reports_backend_features_and_windows_skill(monkeypatch): |
| 1151 | module = _load_computer_use_remote_tool(monkeypatch) |
| 1152 | tool = object.__new__(module.ComputerUseRemote) |
| 1153 | |
| 1154 | message = tool._extract_result( |
| 1155 | "start_session", |
| 1156 | { |
| 1157 | "ok": True, |
| 1158 | "result": { |
| 1159 | "session_id": "s1", |
| 1160 | "width": 3840, |
| 1161 | "height": 2160, |
| 1162 | "backend_id": "windows", |
| 1163 | "backend_family": "windows", |
| 1164 | "features": [ |
| 1165 | "uia-tree-snapshot", |
| 1166 | "uia-structural-targeting", |
| 1167 | ], |
| 1168 | }, |
| 1169 | }, |
| 1170 | ) |
| 1171 | |
| 1172 | assert "session_id=s1" in message |
| 1173 | assert "backend=windows/windows" in message |
| 1174 | assert "features=uia-tree-snapshot, uia-structural-targeting" in message |
| 1175 | assert "host-computer-use-windows" in message |
| 1176 | |
| 1177 | |
| 1178 | def test_computer_use_remote_forwards_linux_window_scope_and_type_guard(monkeypatch): |
| 1179 | module = _load_computer_use_remote_tool(monkeypatch) |
| 1180 | tool = object.__new__(module.ComputerUseRemote) |
| 1181 | window_id = "atspi-pid:57929:path:20.0" |
| 1182 | |
| 1183 | tool.args = { |
| 1184 | "action": "ax_snapshot", |
| 1185 | "pid": 57929, |
| 1186 | "window_id": window_id, |
| 1187 | "max_depth": 4, |
| 1188 | "max_nodes": 80, |
| 1189 | } |
| 1190 | snapshot = tool._build_payload(op_id="op-snapshot", context_id="ctx", action="ax_snapshot") |
| 1191 | tool.args = {"action": "type", "window_id": window_id, "text": "hello", "submit": True} |
| 1192 | typed = tool._build_payload(op_id="op-type", context_id="ctx", action="type") |
| 1193 | |
| 1194 | assert snapshot["pid"] == 57929 |
| 1195 | assert snapshot["window_id"] == window_id |
| 1196 | assert snapshot["max_depth"] == 4 |
| 1197 | assert snapshot["max_nodes"] == 80 |
| 1198 | assert typed["window_id"] == window_id |
| 1199 | assert typed["text"] == "hello" |
| 1200 | assert typed["submit"] is True |
| 1201 | |
| 1202 | |
| 1203 | def test_computer_use_remote_receipts_separate_injection_from_verified_result(monkeypatch): |
| 1204 | module = _load_computer_use_remote_tool(monkeypatch) |
| 1205 | tool = object.__new__(module.ComputerUseRemote) |
| 1206 | window_id = "atspi-pid:57929:path:20.0" |
| 1207 | |
| 1208 | verified = tool._extract_result( |
| 1209 | "type", |
| 1210 | { |
| 1211 | "ok": True, |
| 1212 | "result": { |
| 1213 | "text": "hello", |
| 1214 | "window_id": window_id, |
| 1215 | "focus_verified": True, |
| 1216 | }, |
| 1217 | }, |
| 1218 | ) |
| 1219 | unverified = tool._extract_result( |
| 1220 | "type", |
| 1221 | {"ok": True, "result": {"text": "hello"}}, |
| 1222 | ) |
| 1223 | focused = tool._extract_result( |
| 1224 | "element_action", |
| 1225 | { |
| 1226 | "ok": True, |
| 1227 | "result": { |
| 1228 | "operation": "focus", |
| 1229 | "target": {"element_index": 0, "role": "frame", "title": "Discord"}, |
| 1230 | "requested_dispatch": "foreground", |
| 1231 | "actual_dispatch": "foreground", |
| 1232 | "focus_verified": True, |
| 1233 | }, |
| 1234 | }, |
| 1235 | ) |
| 1236 | scoped = tool._extract_result( |
| 1237 | "ax_snapshot", |
| 1238 | { |
| 1239 | "ok": True, |
| 1240 | "result": { |
| 1241 | "app": {"name": "Discord"}, |
| 1242 | "tree": {"role": "frame", "title": "Discord"}, |
| 1243 | "node_count": 2, |
| 1244 | "window_id": window_id, |
| 1245 | "scoped": True, |
| 1246 | }, |
| 1247 | }, |
| 1248 | ) |
| 1249 | |
| 1250 | assert f"verified active window_id={window_id}" in verified |
| 1251 | assert "destination was not verified" in unverified |
| 1252 | assert "focus_verified=true" in focused |
| 1253 | assert f"scoped to window_id={window_id}" in scoped |
| 1254 | |
| 1255 | |
| 1256 | def test_computer_use_remote_capture_artifact_is_chat_scoped(monkeypatch, tmp_path: Path): |
| 1257 | module = _load_computer_use_remote_tool(monkeypatch) |
| 1258 | |
| 1259 | def fake_get_abs_path(*parts): |
| 1260 | return str(tmp_path.joinpath(*parts)) |
| 1261 | |
| 1262 | def fake_normalize_a0_path(path): |
| 1263 | return "/a0/" + str(Path(path).relative_to(tmp_path)).replace("\\", "/") |
| 1264 | |
| 1265 | monkeypatch.setattr(module.chat_media.files, "get_abs_path", fake_get_abs_path) |
| 1266 | monkeypatch.setattr(module.chat_media.files, "normalize_a0_path", fake_normalize_a0_path) |
| 1267 | |
| 1268 | tool = object.__new__(module.ComputerUseRemote) |
| 1269 | tool.agent = types.SimpleNamespace(context=types.SimpleNamespace(id="ctx-computer")) |
| 1270 | |
| 1271 | display_ref, capture_id = tool._resolve_capture_ref( |
| 1272 | { |
| 1273 | "artifact": { |
| 1274 | "filename": "capture.png", |
| 1275 | "mime": "image/png", |
| 1276 | "encoding": "base64", |
| 1277 | "data": "ZmFrZQ==", |
| 1278 | }, |
| 1279 | } |
| 1280 | ) |
| 1281 | |
| 1282 | assert display_ref.startswith("/a0/usr/chats/ctx-computer/screenshots/computer-use/capture-") |
| 1283 | stored_path = tmp_path / display_ref.removeprefix("/a0/") |
| 1284 | assert stored_path.read_bytes() == b"fake" |
| 1285 | assert capture_id == stored_path.stem |