Add runtime capability scaffolding and align docs

This commit is contained in:
Thales Maciel 2026-03-05 22:57:09 -03:00
parent fb8b985049
commit cbf212bb7b
19 changed files with 1048 additions and 71 deletions

View file

@ -94,7 +94,7 @@ def _stepwise_model_response(payload: dict[str, Any], step: int) -> dict[str, An
"arguments": json.dumps(
{
"vm_id": vm_id,
"command": "printf 'git version 2.44.0\\n'",
"command": "printf 'true\\n'",
}
),
},
@ -125,14 +125,14 @@ def test_run_ollama_tool_demo_happy_path(monkeypatch: pytest.MonkeyPatch) -> Non
result = ollama_demo.run_ollama_tool_demo(log=logs.append)
assert result["fallback_used"] is False
assert "git version" in str(result["exec_result"]["stdout"])
assert str(result["exec_result"]["stdout"]).strip() == "true"
assert result["final_response"] == "Executed git command in ephemeral VM."
assert len(result["tool_events"]) == 4
assert any("[model] input user:" in line for line in logs)
assert any("[model] output assistant:" in line for line in logs)
assert any(line == "[model] input user" for line in logs)
assert any(line == "[model] output assistant" for line in logs)
assert any("[model] tool_call vm_exec" in line for line in logs)
assert any("[tool] calling vm_exec" in line for line in logs)
assert any("[tool] result vm_exec " in line for line in logs)
assert any(line == "[tool] calling vm_exec" for line in logs)
assert any(line == "[tool] result vm_exec" for line in logs)
def test_run_ollama_tool_demo_recovers_from_bad_vm_id(
@ -158,7 +158,7 @@ def test_run_ollama_tool_demo_recovers_from_bad_vm_id(
"arguments": json.dumps(
{
"vm_id": "vm_list_profiles",
"command": "git --version",
"command": ollama_demo.NETWORK_PROOF_COMMAND,
}
),
},
@ -219,7 +219,7 @@ def test_run_ollama_tool_demo_resolves_vm_id_placeholder(
"arguments": json.dumps(
{
"vm_id": "<vm_id_returned_by_vm_create>",
"command": "printf 'git version 2.44.0\\n'",
"command": "printf 'true\\n'",
"timeout_seconds": "300",
}
),
@ -292,14 +292,49 @@ def test_run_ollama_tool_demo_uses_fallback_when_not_strict(
monkeypatch.setattr(ollama_demo, "_post_chat_completion", fake_post_chat_completion)
monkeypatch.setattr(ollama_demo, "VmManager", TestVmManager)
monkeypatch.setattr(
ollama_demo,
"_run_direct_lifecycle_fallback",
lambda manager: {
"vm_id": "vm-1",
"command": ollama_demo.NETWORK_PROOF_COMMAND,
"stdout": "true\n",
"stderr": "",
"exit_code": 0,
"duration_ms": 5,
"execution_mode": "host_compat",
"cleanup": {"deleted": True, "reason": "post_exec_cleanup", "vm_id": "vm-1"},
},
)
logs: list[str] = []
result = ollama_demo.run_ollama_tool_demo(strict=False, log=logs.append)
assert result["fallback_used"] is True
assert int(result["exec_result"]["exit_code"]) == 0
assert any("[model] output assistant: No tools" in line for line in logs)
assert any(line == "[model] output assistant" for line in logs)
assert any("[fallback]" in line for line in logs)
def test_run_ollama_tool_demo_verbose_logs_values(monkeypatch: pytest.MonkeyPatch) -> None:
requests = 0
def fake_post_chat_completion(base_url: str, payload: dict[str, Any]) -> dict[str, Any]:
del base_url
nonlocal requests
requests += 1
return _stepwise_model_response(payload, requests)
monkeypatch.setattr(ollama_demo, "_post_chat_completion", fake_post_chat_completion)
logs: list[str] = []
result = ollama_demo.run_ollama_tool_demo(verbose=True, log=logs.append)
assert result["fallback_used"] is False
assert str(result["exec_result"]["stdout"]).strip() == "true"
assert any("[model] input user:" in line for line in logs)
assert any("[model] tool_call vm_list_profiles args={}" in line for line in logs)
assert any("[tool] result vm_exec " in line for line in logs)
@pytest.mark.parametrize(
("tool_call", "error"),
[
@ -346,8 +381,8 @@ def test_run_ollama_tool_demo_max_rounds(monkeypatch: pytest.MonkeyPatch) -> Non
("exec_result", "error"),
[
("bad", "result shape is invalid"),
({"exit_code": 1, "stdout": "git version 2"}, "expected exit_code=0"),
({"exit_code": 0, "stdout": "no git"}, "did not contain `git version`"),
({"exit_code": 1, "stdout": "true"}, "expected exit_code=0"),
({"exit_code": 0, "stdout": "false"}, "did not confirm repository clone success"),
],
)
def test_run_ollama_tool_demo_exec_result_validation(
@ -404,7 +439,7 @@ def test_dispatch_tool_call_coverage(tmp_path: Path) -> None:
executed = ollama_demo._dispatch_tool_call(
manager,
"vm_exec",
{"vm_id": vm_id, "command": "printf 'git version\\n'", "timeout_seconds": "30"},
{"vm_id": vm_id, "command": "printf 'true\\n'", "timeout_seconds": "30"},
)
assert int(executed["exit_code"]) == 0
with pytest.raises(RuntimeError, match="unexpected tool requested by model"):
@ -529,6 +564,13 @@ def test_build_parser_defaults() -> None:
args = parser.parse_args([])
assert args.model == ollama_demo.DEFAULT_OLLAMA_MODEL
assert args.base_url == ollama_demo.DEFAULT_OLLAMA_BASE_URL
assert args.verbose is False
def test_build_parser_verbose_flag() -> None:
parser = ollama_demo._build_parser()
args = parser.parse_args(["-v"])
assert args.verbose is True
def test_main_uses_parser_and_prints_logs(
@ -537,21 +579,51 @@ def test_main_uses_parser_and_prints_logs(
) -> None:
class StubParser:
def parse_args(self) -> argparse.Namespace:
return argparse.Namespace(base_url="http://x", model="m")
return argparse.Namespace(base_url="http://x", model="m", verbose=False)
monkeypatch.setattr(ollama_demo, "_build_parser", lambda: StubParser())
monkeypatch.setattr(
ollama_demo,
"run_ollama_tool_demo",
lambda base_url, model, strict=True, log=None: {
"exec_result": {"exit_code": 0, "stdout": "git version 2.44.0\n"},
lambda base_url, model, strict=True, verbose=False, log=None: {
"exec_result": {
"exit_code": 0,
"stdout": "true\n",
"execution_mode": "host_compat",
},
"fallback_used": False,
},
)
ollama_demo.main()
output = capsys.readouterr().out
assert "[summary] exit_code=0 fallback_used=False" in output
assert "[summary] stdout=git version 2.44.0" in output
assert "[summary] exit_code=0 fallback_used=False execution_mode=host_compat" in output
assert "[summary] stdout=" not in output
def test_main_verbose_prints_stdout(
monkeypatch: pytest.MonkeyPatch,
capsys: pytest.CaptureFixture[str],
) -> None:
class StubParser:
def parse_args(self) -> argparse.Namespace:
return argparse.Namespace(base_url="http://x", model="m", verbose=True)
monkeypatch.setattr(ollama_demo, "_build_parser", lambda: StubParser())
monkeypatch.setattr(
ollama_demo,
"run_ollama_tool_demo",
lambda base_url, model, strict=True, verbose=False, log=None: {
"exec_result": {
"exit_code": 0,
"stdout": "true\n",
"execution_mode": "host_compat",
},
"fallback_used": False,
},
)
ollama_demo.main()
output = capsys.readouterr().out
assert "[summary] stdout=true" in output
def test_main_logs_error_and_exits_nonzero(
@ -560,12 +632,18 @@ def test_main_logs_error_and_exits_nonzero(
) -> None:
class StubParser:
def parse_args(self) -> argparse.Namespace:
return argparse.Namespace(base_url="http://x", model="m")
return argparse.Namespace(base_url="http://x", model="m", verbose=False)
monkeypatch.setattr(ollama_demo, "_build_parser", lambda: StubParser())
def fake_run(base_url: str, model: str, strict: bool = True, log: Any = None) -> dict[str, Any]:
del base_url, model, strict, log
def fake_run(
base_url: str,
model: str,
strict: bool = True,
verbose: bool = False,
log: Any = None,
) -> dict[str, Any]:
del base_url, model, strict, verbose, log
raise RuntimeError("demo did not execute a successful vm_exec")
monkeypatch.setattr(ollama_demo, "run_ollama_tool_demo", fake_run)