hermes-agent/tests/tools/test_terminal_task_cwd.py
ethernet 7e84d2b5a4 fix(terminal): ignore stale env.cwd from a different session's cd
The terminal environment is shared process-globally (collapsed to the
default key), so env.cwd tracks the LAST session that ran a command.
_resolve_command_cwd() trusted env.cwd unconditionally — no ownership
check — so when session A left env.cwd pointing at A's checkout,
session B's first terminal command inherited A's stale cwd and ran in
the wrong workspace.

The file tools already solved this exact shared-env problem with
_live_cwd_if_owned() checking env.cwd_owner. The terminal tool never
got the same guard.

Fix: capture env.cwd_owner BEFORE the current session claims it, and
pass it as prev_owner to _resolve_command_cwd. When the previous owner
was a different session, env.cwd is stale — fall through to default_cwd
(the config/override cwd for this session) instead. Once the session
has claimed the env, subsequent calls in the same session still trust
env.cwd so in-session  state survives.
2026-07-14 15:31:39 -04:00

330 lines
12 KiB
Python

"""Regression tests for task/session cwd propagation in terminal_tool."""
import json
from types import SimpleNamespace
import tools.terminal_tool as terminal_tool
def _minimal_terminal_config(cwd="/default"):
return {
"env_type": "local",
"cwd": cwd,
"timeout": 60,
"lifetime_seconds": 3600,
}
def test_foreground_command_uses_registered_task_cwd_for_existing_environment(monkeypatch):
"""ACP can update task cwd after the local env exists; foreground must honor it."""
calls = []
class FakeEnv:
env = {}
def execute(self, command, **kwargs):
calls.append((command, kwargs))
return {"output": "ok", "returncode": 0}
task_id = "acp-session-1"
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: FakeEnv()})
monkeypatch.setattr(terminal_tool, "_last_activity", {})
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {task_id: {"cwd": "/workspace/acp"}})
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config())
monkeypatch.setattr(
terminal_tool,
"_check_all_guards",
lambda command, env_type, **kwargs: {"approved": True},
)
result = json.loads(terminal_tool.terminal_tool(command="pwd", task_id=task_id))
assert result["exit_code"] == 0
assert calls == [("pwd", {"timeout": 60, "cwd": "/workspace/acp"})]
def test_explicit_workdir_still_wins_over_registered_task_cwd(monkeypatch):
calls = []
class FakeEnv:
env = {}
def execute(self, command, **kwargs):
calls.append(kwargs)
return {"output": "ok", "returncode": 0}
task_id = "acp-session-1"
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: FakeEnv()})
monkeypatch.setattr(terminal_tool, "_last_activity", {})
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {task_id: {"cwd": "/workspace/acp"}})
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config())
monkeypatch.setattr(
terminal_tool,
"_check_all_guards",
lambda command, env_type, **kwargs: {"approved": True},
)
result = json.loads(
terminal_tool.terminal_tool(
command="pwd",
task_id=task_id,
workdir="/explicit/workdir",
)
)
assert result["exit_code"] == 0
assert calls == [{"timeout": 60, "cwd": "/explicit/workdir"}]
def test_foreground_command_prefers_live_env_cwd_over_init_time_cwd(monkeypatch):
"""A prior `cd` updates env.cwd; terminal_tool must honor that live cwd."""
calls = []
class FakeEnv:
env = {}
cwd = "/workspace/live"
def execute(self, command, **kwargs):
calls.append((command, kwargs))
return {"output": "ok", "returncode": 0}
task_id = "session-live-cwd"
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: FakeEnv()})
monkeypatch.setattr(terminal_tool, "_last_activity", {})
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {task_id: {"cwd": "/workspace/init"}})
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config(cwd="/workspace/init"))
monkeypatch.setattr(terminal_tool, "_start_cleanup_thread", lambda: None)
monkeypatch.setattr(terminal_tool, "_resolve_container_task_id", lambda value: value or "default")
monkeypatch.setattr(
terminal_tool,
"_check_all_guards",
lambda command, env_type, **kwargs: {"approved": True},
)
result = json.loads(terminal_tool.terminal_tool(command="pwd", task_id=task_id))
assert result["exit_code"] == 0
assert calls == [("pwd", {"timeout": 60, "cwd": "/workspace/live"})]
def test_background_command_prefers_live_env_cwd_over_init_time_cwd(monkeypatch):
"""Background process launches must also use the live session cwd."""
class FakeEnv:
env = {}
cwd = "/workspace/live"
class FakeRegistry:
def __init__(self):
self.calls = []
self.pending_watchers = []
def spawn_local(self, **kwargs):
self.calls.append(kwargs)
return SimpleNamespace(id="proc_test", pid=1234)
import tools.process_registry as process_registry_mod
registry = FakeRegistry()
task_id = "session-live-cwd-bg"
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: FakeEnv()})
monkeypatch.setattr(terminal_tool, "_last_activity", {})
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {task_id: {"cwd": "/workspace/init"}})
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config(cwd="/workspace/init"))
monkeypatch.setattr(terminal_tool, "_start_cleanup_thread", lambda: None)
monkeypatch.setattr(terminal_tool, "_resolve_container_task_id", lambda value: value or "default")
monkeypatch.setattr(
terminal_tool,
"_check_all_guards",
lambda command, env_type, **kwargs: {"approved": True},
)
monkeypatch.setattr(process_registry_mod, "process_registry", registry)
result = json.loads(
terminal_tool.terminal_tool(
command="sleep 1",
task_id=task_id,
background=True,
)
)
assert result["exit_code"] == 0
# session_key falls back to the raw task_id when no gateway contextvar is set
# (it doesn't propagate to tool-worker threads), so process.kill / stop can
# still find and terminate this background process.
assert registry.calls == [{
"command": "sleep 1",
"cwd": "/workspace/live",
"task_id": task_id,
"session_key": task_id,
"env_vars": {},
"use_pty": False,
}]
def test_registering_cwd_override_updates_live_env_cwd(monkeypatch):
"""An ACP ``update_cwd`` (re-)registered mid-session must win over a
previously ``cd``-ed live ``env.cwd``.
Preferring live ``env.cwd`` (so session-local ``cd`` survives) means a
freshly registered ``cwd`` override would otherwise sit *below* the
already-set ``env.cwd`` and be silently ignored. ``register_task_env_overrides``
syncs the new cwd onto the live cached env so an explicit ACP project-root
change takes effect, as the editor client expects.
"""
class FakeEnv:
env = {}
cwd = "/workspace/old"
task_id = "acp-session-update"
fake_env = FakeEnv()
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: fake_env})
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {})
terminal_tool.register_task_env_overrides(task_id, {"cwd": "/workspace/new"})
# The live env now reflects the editor's new project root.
assert fake_env.cwd == "/workspace/new"
# A subsequent command resolves to the new cwd (env.cwd precedence).
assert terminal_tool._resolve_command_cwd(
workdir=None, env=fake_env, default_cwd="/workspace/config"
) == "/workspace/new"
def test_registering_cwd_override_noop_when_no_live_env(monkeypatch):
"""Registering an override before the env exists must not crash; the cwd
is applied at env creation time instead."""
monkeypatch.setattr(terminal_tool, "_active_environments", {})
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {})
# Should not raise even though no env is cached yet.
terminal_tool.register_task_env_overrides("acp-session-pending", {"cwd": "/workspace/new"})
assert terminal_tool._task_env_overrides["acp-session-pending"] == {"cwd": "/workspace/new"}
def test_registering_non_cwd_override_leaves_live_env_cwd_untouched(monkeypatch):
"""A non-cwd override (e.g. a per-task Modal image) must not disturb the
live env's cwd."""
class FakeEnv:
env = {}
cwd = "/workspace/keep"
task_id = "rl-rollout-1"
fake_env = FakeEnv()
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: fake_env})
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {})
terminal_tool.register_task_env_overrides(task_id, {"modal_image": "custom:latest"})
assert fake_env.cwd == "/workspace/keep"
def test_stale_env_cwd_from_different_session_is_ignored(monkeypatch):
"""A different session's `cd` left env.cwd pointing at its checkout.
The terminal env is shared (collapsed to "default"), so env.cwd tracks the
LAST session that ran a command. When session B claims the env after
session A left it in A's worktree, the first command must NOT run in A's
leftover cwd — it must fall through to the config/override cwd (this
session's own workspace).
"""
calls = []
class FakeEnv:
env = {}
cwd = "/home/user/src/hermes-desktop-tipc/apps/desktop"
cwd_owner = "session-A-key"
def execute(self, command, **kwargs):
calls.append((command, kwargs))
return {"output": "ok", "returncode": 0}
task_id = "session-B"
monkeypatch.setattr(terminal_tool, "_active_environments", {"default": FakeEnv()})
monkeypatch.setattr(terminal_tool, "_last_activity", {})
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {})
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config(cwd="/home/user/src/hermes-agent"))
monkeypatch.setattr(terminal_tool, "_start_cleanup_thread", lambda: None)
monkeypatch.setattr(terminal_tool, "_resolve_container_task_id", lambda value: "default")
monkeypatch.setattr(
terminal_tool,
"_check_all_guards",
lambda command, env_type, **kwargs: {"approved": True},
)
result = json.loads(terminal_tool.terminal_tool(command="pwd", task_id=task_id))
assert result["exit_code"] == 0
# The command must run in the config cwd (hermes-agent), NOT the stale
# env.cwd left by session A (hermes-desktop-tipc).
assert calls == [("pwd", {"timeout": 60, "cwd": "/home/user/src/hermes-agent"})]
def test_same_session_env_cwd_is_trusted_after_first_claim(monkeypatch):
"""Once a session has claimed the env, subsequent commands trust env.cwd.
The prev_owner check only rejects env.cwd when a DIFFERENT session owned it
before this call. After the first command (which claims ownership),
subsequent calls in the same session should trust the live env.cwd so that
in-session `cd` state survives.
"""
calls = []
class FakeEnv:
env = {}
cwd = "/workspace/deep"
cwd_owner = "session-X"
def execute(self, command, **kwargs):
calls.append((command, kwargs))
return {"output": "ok", "returncode": 0}
env = FakeEnv()
task_id = "session-X"
monkeypatch.setattr(terminal_tool, "_active_environments", {"default": env})
monkeypatch.setattr(terminal_tool, "_last_activity", {})
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {})
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config(cwd="/workspace/config"))
monkeypatch.setattr(terminal_tool, "_start_cleanup_thread", lambda: None)
monkeypatch.setattr(terminal_tool, "_resolve_container_task_id", lambda value: "default")
monkeypatch.setattr(
terminal_tool,
"_check_all_guards",
lambda command, env_type, **kwargs: {"approved": True},
)
# First call: env was owned by "session-X" (same session_key since
# get_current_session_key falls back to task_id). prev_owner == current
# session, so env.cwd is trusted.
result = json.loads(terminal_tool.terminal_tool(command="pwd", task_id=task_id))
assert result["exit_code"] == 0
assert calls == [("pwd", {"timeout": 60, "cwd": "/workspace/deep"})]
def test_safe_getcwd_returns_real_cwd(monkeypatch):
monkeypatch.setattr(terminal_tool.os, "getcwd", lambda: "/home/user/project")
assert terminal_tool._safe_getcwd() == "/home/user/project"
def test_safe_getcwd_falls_back_to_terminal_cwd_when_cwd_deleted(monkeypatch):
def _boom():
raise FileNotFoundError("[Errno 2] No such file or directory")
monkeypatch.setattr(terminal_tool.os, "getcwd", _boom)
monkeypatch.setenv("TERMINAL_CWD", "/srv/work")
assert terminal_tool._safe_getcwd() == "/srv/work"
def test_safe_getcwd_falls_back_to_home_when_no_terminal_cwd(monkeypatch):
def _boom():
raise FileNotFoundError()
monkeypatch.setattr(terminal_tool.os, "getcwd", _boom)
monkeypatch.delenv("TERMINAL_CWD", raising=False)
monkeypatch.setattr(terminal_tool.os.path, "expanduser", lambda p: "/home/me")
assert terminal_tool._safe_getcwd() == "/home/me"