Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 13 additions & 2 deletions src/agent_env/task_step/task_steps/deploy_env.py
Original file line number Diff line number Diff line change
Expand Up @@ -77,6 +77,16 @@ def run_state(self, user_overrides: Mapping[str, Any]) -> tuple[str | None, str
return (user_overrides.get("env_state_type") or self.env_state_type,
user_overrides.get("env_state_instance_id") or self.env_state_instance_id)

def run_env_version(self, context: TaskStepContext) -> Optional[int]:
"""The env version a run deploys: this step's ``env_version`` step override, over the stored one."""
overrides = self.step_param_overrides(context)
if "env_version" not in overrides:
return self.env_version
version = overrides["env_version"]
if not isinstance(version, int) or isinstance(version, bool):
raise ValueError(f"deploy_env '{self.id}': step override env_version must be an integer, got {version!r}")
return version

def to_dict(self) -> dict:
base = super().to_dict()
base["env_id"] = self.env_id
Expand Down Expand Up @@ -157,15 +167,16 @@ async def execute(self, context: TaskStepContext) -> TaskStepContext:
if existing is not None:
raise RuntimeError(f"Env '{self.env_id}' is already deployed")

env = Env.get(self.env_id, self.env_version)
env_version = self.run_env_version(context)
env = Env.get(self.env_id, env_version)
user_overrides = context.metadata.get("user_overrides", {})
resolved_sandbox_type = user_overrides.get("env_sandbox") or self.sandbox_type
resolved_env_state_type, resolved_env_state_instance_id = self.run_state(user_overrides)
_ttl_override = user_overrides.get("ttl_seconds")
resolved_ttl = _ttl_override if _ttl_override is not None else self.ttl_seconds
attribution = deploy_attribution(self, context)
logger.info(
f"Deploying env '{self.env_id}' (ttl={resolved_ttl}s, "
f"Deploying env '{self.env_id}' (version={env_version}, ttl={resolved_ttl}s, "
f"disk_size_gb={self.disk_size_gb}, gateway_mode={self.gateway_mode}, "
f"cpu={self.cpu}, memory_mb={self.memory_mb}, "
f"sandbox_type={resolved_sandbox_type}, env_state_type={resolved_env_state_type}, "
Expand Down
22 changes: 18 additions & 4 deletions src/agent_env/task_step/task_steps/modify_env_tool_access.py
Original file line number Diff line number Diff line change
Expand Up @@ -34,7 +34,7 @@ def __init__(
raise ValueError(f"action must be {TOOL_DISABLE_ACTION!r} or {TOOL_ENABLE_ACTION!r}, got {action!r}")
if not isinstance(role, str) or not role:
raise ValueError("role must be a non-empty string")
if not isinstance(tools, list) or not all(isinstance(t, str) for t in tools):
if not _is_tool_list(tools):
raise ValueError("tools must be a list of strings")
self.env_id = env_id
self.action = action
Expand All @@ -59,22 +59,36 @@ def from_dict(cls, data: dict) -> ModifyEnvToolAccessStep:
tools=data["tools"],
)

def run_tools(self, context: TaskStepContext) -> list[str]:
"""The tools a run changes: this step's ``tools`` step override, which replaces the stored list."""
overrides = self.step_param_overrides(context)
if "tools" not in overrides:
return self.tools
if not _is_tool_list(tools := overrides["tools"]):
raise ValueError(f"modify_env_tool_access '{self.id}': step override tools must be a list of strings, got {tools!r}")
return tools

async def execute(self, context: TaskStepContext) -> TaskStepContext:
deployed = next((d for d in context.deployed_envs if d.env_id == self.env_id), None)
if deployed is None:
raise RuntimeError(f"Env '{self.env_id}' not found in context.deployed_envs")
body = await deployed.invoke(_TOOL_ACCESS_URIS[self.action], self.action, {"role": self.role, "tools": self.tools})
logger.info(f"{self.action} role={self.role!r} tools={self.tools} on env={self.env_id}: {body}")
tools = self.run_tools(context)
body = await deployed.invoke(_TOOL_ACCESS_URIS[self.action], self.action, {"role": self.role, "tools": tools})
logger.info(f"{self.action} role={self.role!r} tools={tools} on env={self.env_id}: {body}")
context.metadata.setdefault("tool_access_changes", []).append({
"step_id": self.id,
"env_id": self.env_id,
"action": self.action,
"role": self.role,
"tools_requested": self.tools,
"tools_requested": tools,
"role_state_after": {"disabled": body.get("disabled", []), "allowed": body.get("allowed", [])},
})
return context


def _is_tool_list(tools) -> bool:
return isinstance(tools, list) and all(isinstance(t, str) for t in tools)


# The gateway extension each action is advertised under on the env card.
_TOOL_ACCESS_URIS = {TOOL_DISABLE_ACTION: EXT_DISABLE_TOOL_URI, TOOL_ENABLE_ACTION: EXT_ENABLE_TOOL_URI}
31 changes: 31 additions & 0 deletions tst/unit/task_step/deploy_env_test.py
Original file line number Diff line number Diff line change
Expand Up @@ -555,3 +555,34 @@ async def test_execute_refuses_child_envs_for_an_env_whose_deploy_takes_none():
with patch("agent_env.env.env.Env") as Env, pytest.raises(ValueError, match="have no child envs"):
Env.get.return_value = env
await step.execute(TaskStepContext())


def _step_overrides(step_id: str, params: dict) -> dict:
return {"user_overrides": {"step_params": {step_id: params}}}


@pytest.mark.asyncio
@pytest.mark.parametrize("metadata, expected", [
({}, 2),
(_step_overrides("t-1.deploy_env", {"env_version": 5}), 5),
(_step_overrides("another-step", {"env_version": 5}), 2),
])
async def test_execute_deploys_the_env_version_a_step_override_names(metadata, expected):
step = DeployEnvTaskStep(id="t-1.deploy_env", version=None, env_id="art-1", env_version=2)

with patch("agent_env.env.env.Env") as Env:
Env.get.return_value = _fake_env(_deployed_env(None))
await step.execute(TaskStepContext(metadata=metadata))

Env.get.assert_called_once_with("art-1", expected)
assert step.env_version == 2
Comment on lines +573 to +578

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Recorded version goes unchecked

The override case requests version 5, but _deployed_env(None) always returns a deployment with env_version=2. The test checks only the Env.get arguments, so it passes even when the returned context records the wrong version. Make the fake deployment match the selected version and assert the version in the returned context. This protects the record that later verifiers rely on.

Prompt To Fix With AI
This is a comment left during a code review.
Path: tst/unit/task_step/deploy_env_test.py
Line: 573-578

Comment:
**Recorded version goes unchecked**

The override case requests version 5, but `_deployed_env(None)` always returns a deployment with `env_version=2`. The test checks only the `Env.get` arguments, so it passes even when the returned context records the wrong version. Make the fake deployment match the selected version and assert the version in the returned context. This protects the record that later verifiers rely on.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Fix in Cursor Fix in Claude Code Fix in Codex



@pytest.mark.asyncio
@pytest.mark.parametrize("bad", ["5", 5.0, True, None])
async def test_execute_rejects_an_env_version_override_that_is_not_an_integer(bad):
step = DeployEnvTaskStep(id="t-1.deploy_env", version=None, env_id="art-1", env_version=2)

with patch("agent_env.env.env.Env") as Env, pytest.raises(ValueError, match="env_version must be an integer"):
await step.execute(TaskStepContext(metadata=_step_overrides("t-1.deploy_env", {"env_version": bad})))
Env.get.assert_not_called()
31 changes: 31 additions & 0 deletions tst/unit/task_step/test_modify_env_tool_access.py
Original file line number Diff line number Diff line change
Expand Up @@ -51,6 +51,37 @@ async def test_a_gateway_error_raises(monkeypatch):
await _step("disable").execute(TaskStepContext(deployed_envs=[_record()], metadata={}))


@pytest.mark.asyncio
@pytest.mark.parametrize("step_params, expected", [
({}, ["slack_send"]),
({"tools-1": {"tools": ["mail_send", "mail_delete"]}}, ["mail_send", "mail_delete"]),
({"tools-1": {"tools": []}}, []),
({"other-step": {"tools": ["mail_send"]}}, ["slack_send"]),
])
async def test_a_tools_step_override_replaces_the_stored_tools(monkeypatch, step_params, expected):
sent = _mock_gateway(monkeypatch, {"ok": True, "disabled": expected, "allowed": []})
ctx = TaskStepContext(deployed_envs=[_record()], metadata={"user_overrides": {"step_params": step_params}})
step = _step("disable")

out = await step.execute(ctx)

[request] = sent
assert json.loads(request.content) == {"role": "default", "tools": expected}
assert out.metadata["tool_access_changes"][0]["tools_requested"] == expected
assert step.tools == ["slack_send"]


@pytest.mark.asyncio
@pytest.mark.parametrize("bad", ["mail_send", ["mail_send", 3], None, {"mail_send": True}])
async def test_a_tools_override_that_is_not_a_list_of_strings_raises_before_any_request(monkeypatch, bad):
sent = _mock_gateway(monkeypatch, {})
ctx = TaskStepContext(deployed_envs=[_record()], metadata={"user_overrides": {"step_params": {"tools-1": {"tools": bad}}}})

with pytest.raises(ValueError, match="step override tools must be a list of strings"):
await _step("disable").execute(ctx)
assert sent == []


def _step(action: str) -> ModifyEnvToolAccessStep:
return ModifyEnvToolAccessStep(id="tools-1", version=None, env_id="env-x", action=action, role="default",
tools=["slack_send"])
Expand Down
Loading