Files
hermes-agent/tests/gateway/test_profile_isolation_runtime.py
ethernet 890bbbda1f Merge remote-tracking branch 'origin/main' into ethie/pm-clean
# Conflicts:
#	apps/desktop/e2e/archived-hidden-session-recoverable.spec.ts
#	apps/desktop/e2e/bot-chat-message-agent-friendly-name.spec.ts
#	apps/desktop/e2e/bot-mailbox-unreadable-ticket.spec.ts
#	apps/desktop/e2e/bot-mode-roster-localized.spec.ts
#	apps/desktop/e2e/bot-mode-row-click-mirrors-registry.spec.ts
#	apps/desktop/e2e/bot-mode-tab-shows-bot-name.spec.ts
#	apps/desktop/e2e/bot-roster-group-row-organisation.spec.ts
#	apps/desktop/e2e/bot-roster-ignores-infra-dirs.spec.ts
#	apps/desktop/e2e/bot-roster-timestamp-meta.spec.ts
#	apps/desktop/e2e/bot-roster-user-sections.spec.ts
#	apps/desktop/e2e/bot-routines-pane-narrow.spec.ts
#	apps/desktop/e2e/bot-row-open-recent-session.spec.ts
#	apps/desktop/e2e/bot-tile-ignores-ambient-composer-model.spec.ts
#	apps/desktop/e2e/group-composer-auto-grow.spec.ts
#	apps/desktop/e2e/group-create-gate-remote-roster.spec.ts
#	apps/desktop/e2e/group-prompt-renamed-primary-handle.spec.ts
#	apps/desktop/e2e/hosted-room-backend-continuity.spec.ts
#	apps/desktop/e2e/hosted-room-legacy-store-migration.spec.ts
#	apps/desktop/e2e/settings-scope-chips-bot-title.spec.ts
#	apps/desktop/e2e/worktree-branch-status.spec.ts
#	apps/desktop/electron/backend-probes.test.ts
#	apps/desktop/electron/connection-apply.test.ts
#	apps/desktop/electron/desktop-electron-pin.test.ts
#	apps/desktop/electron/desktop-uninstall.test.ts
#	apps/desktop/electron/gateway-file-download-transport.test.ts
#	apps/desktop/electron/gateway-stop-before-update.test.ts
#	apps/desktop/electron/github-api-auth.test.ts
#	apps/desktop/electron/registry-primary-profile-scope.test.ts
#	apps/desktop/electron/update-api-check.test.ts
#	apps/desktop/electron/update-handoff-marker.test.ts
#	apps/desktop/electron/venv-blocker-scan.test.ts
#	apps/desktop/scripts/after-extract.test.mjs
#	apps/desktop/scripts/local-pack-publish.test.mjs
#	apps/desktop/scripts/tasks-scroll.test.mjs
#	apps/desktop/src/app/settings/model-settings.test.tsx
#	apps/desktop/src/app/updates-overlay.blockers.test.tsx
#	apps/desktop/src/components/desktop-install-overlay.test.tsx
#	apps/desktop/src/lib/update-copy.test.ts
#	scripts/ci/check_os_marker_fakes.py
#	tests-js/desktop-mac-usage-descriptions.test.ts
#	tests-js/node-engine-alignment.test.ts
#	tests/agent/lsp/test_install_and_lint_fixes.py
#	tests/agent/test_command_token_source.py
#	tests/agent/test_compression_boundary_hook.py
#	tests/agent/test_create_openai_client_ssl_verify.py
#	tests/agent/test_custom_provider_ca_probes.py
#	tests/agent/test_endpoint_blackhole.py
#	tests/agent/test_estimator_parity.py
#	tests/agent/test_in_place_compaction.py
#	tests/agent/test_moa_loop_mode.py
#	tests/agent/test_model_metadata.py
#	tests/agent/test_skill_session_platform_gate.py
#	tests/agent/test_skill_utils.py
#	tests/agent/test_ssl_ca_guard.py
#	tests/computer_use/test_doctor.py
#	tests/cron/test_codex_execution_paths.py
#	tests/cron/test_cron_bot_chat_delivery.py
#	tests/cron/test_cron_script.py
#	tests/cron/test_media_delivery_parity.py
#	tests/cron/test_misfire_catchup.py
#	tests/cron/test_parallel_pool.py
#	tests/cron/test_recurring_eagain_redispatch.py
#	tests/gateway/test_choice_picker.py
#	tests/gateway/test_control_socket_windows_live.py
#	tests/gateway/test_dingtalk.py
#	tests/gateway/test_feishu.py
#	tests/gateway/test_feishu_onboard.py
#	tests/gateway/test_gateway_shutdown.py
#	tests/gateway/test_matrix.py
#	tests/gateway/test_model_command_custom_providers.py
#	tests/gateway/test_reasoning_command.py
#	tests/gateway/test_runtime_footer.py
#	tests/gateway/test_session.py
#	tests/gateway/test_session_hygiene.py
#	tests/gateway/test_status.py
#	tests/gateway/test_teams.py
#	tests/gateway/test_turn_lease.py
#	tests/gateway/test_whatsapp_connect.py
#	tests/hermes_cli/test_approvals_command.py
#	tests/hermes_cli/test_auth_store_lock_concurrent.py
#	tests/hermes_cli/test_backup.py
#	tests/hermes_cli/test_banner_git_state.py
#	tests/hermes_cli/test_certifi_repair.py
#	tests/hermes_cli/test_cmd_update.py
#	tests/hermes_cli/test_compat_manifest_targets.py
#	tests/hermes_cli/test_computer_use_cli.py
#	tests/hermes_cli/test_cpr_local_leak.py
#	tests/hermes_cli/test_dashboard_auth_gate.py
#	tests/hermes_cli/test_dashboard_procs_kill_grace.py
#	tests/hermes_cli/test_desktop_lifecycle_windows_live.py
#	tests/hermes_cli/test_doctor.py
#	tests/hermes_cli/test_doctor_command_install.py
#	tests/hermes_cli/test_fleet_config_migration_windows_live.py
#	tests/hermes_cli/test_gateway.py
#	tests/hermes_cli/test_gateway_platform_gating.py
#	tests/hermes_cli/test_gateway_restart_loop.py
#	tests/hermes_cli/test_gateway_task_probe.py
#	tests/hermes_cli/test_gateway_wsl.py
#	tests/hermes_cli/test_gui_command.py
#	tests/hermes_cli/test_install_cua_driver.py
#	tests/hermes_cli/test_kanban_db.py
#	tests/hermes_cli/test_lazy_command_exports.py
#	tests/hermes_cli/test_lazy_refresh_venv_repair.py
#	tests/hermes_cli/test_linux_desktop_entry.py
#	tests/hermes_cli/test_local_runtime.py
#	tests/hermes_cli/test_local_runtime_updates.py
#	tests/hermes_cli/test_managed_uv.py
#	tests/hermes_cli/test_mcp_reload_confirm_gate.py
#	tests/hermes_cli/test_nous_subscription.py
#	tests/hermes_cli/test_npm_engine.py
#	tests/hermes_cli/test_personality_none.py
#	tests/hermes_cli/test_pet_toggle.py
#	tests/hermes_cli/test_plan_reconciliation_windows_live.py
#	tests/hermes_cli/test_plugin_event_bus.py
#	tests/hermes_cli/test_plugin_manifest_v2.py
#	tests/hermes_cli/test_plugin_packs.py
#	tests/hermes_cli/test_plugins_cmd.py
#	tests/hermes_cli/test_plugins_cmd_enable_disable_nested.py
#	tests/hermes_cli/test_process_identity.py
#	tests/hermes_cli/test_profiles.py
#	tests/hermes_cli/test_profiles_sidebar_cache.py
#	tests/hermes_cli/test_pty_bridge.py
#	tests/hermes_cli/test_resolve_turn_limit.py
#	tests/hermes_cli/test_serve_runtime_inventory.py
#	tests/hermes_cli/test_session_vacuum_config.py
#	tests/hermes_cli/test_set_config_value.py
#	tests/hermes_cli/test_signal_handler_kanban_worker.py
#	tests/hermes_cli/test_slash_confirm_windows.py
#	tests/hermes_cli/test_stale_pid_guard.py
#	tests/hermes_cli/test_startup_fast_guards.py
#	tests/hermes_cli/test_status.py
#	tests/hermes_cli/test_telegram_managed_bot.py
#	tests/hermes_cli/test_tools_config.py
#	tests/hermes_cli/test_update_apply_shallow_count.py
#	tests/hermes_cli/test_update_autostash.py
#	tests/hermes_cli/test_update_concurrent_quarantine.py
#	tests/hermes_cli/test_update_fetch_failure_classifier.py
#	tests/hermes_cli/test_update_fleet_probe_resume_token.py
#	tests/hermes_cli/test_update_handoff_backend_reap.py
#	tests/hermes_cli/test_update_handoff_desktop_rebuild.py
#	tests/hermes_cli/test_update_head_moved_gate.py
#	tests/hermes_cli/test_update_host_obligation.py
#	tests/hermes_cli/test_update_import_guard.py
#	tests/hermes_cli/test_update_interrupted_recovery.py
#	tests/hermes_cli/test_update_inventory.py
#	tests/hermes_cli/test_update_launchd_unloaded_gateway.py
#	tests/hermes_cli/test_update_missing_configured_deps.py
#	tests/hermes_cli/test_update_modified_notice.py
#	tests/hermes_cli/test_update_multiplex_migration_hook.py
#	tests/hermes_cli/test_update_no_gateway_restart.py
#	tests/hermes_cli/test_update_orphan_backend_reap.py
#	tests/hermes_cli/test_update_parked_branch_guard.py
#	tests/hermes_cli/test_update_post_pull_syntax_guard.py
#	tests/hermes_cli/test_update_receipt.py
#	tests/hermes_cli/test_update_self_lock.py
#	tests/hermes_cli/test_update_shim_fail_closed.py
#	tests/hermes_cli/test_update_shim_self_lock.py
#	tests/hermes_cli/test_update_sqlite_remediation.py
#	tests/hermes_cli/test_update_stale_dashboard.py
#	tests/hermes_cli/test_update_stale_virtualenv.py
#	tests/hermes_cli/test_update_venv_health.py
#	tests/hermes_cli/test_update_venv_ownership_preflight.py
#	tests/hermes_cli/test_update_wedged_gateway.py
#	tests/hermes_cli/test_update_yes_flag.py
#	tests/hermes_cli/test_update_zip_two_phase.py
#	tests/hermes_cli/test_urllib_security.py
#	tests/hermes_cli/test_ux_messages_auth_config.py
#	tests/hermes_cli/test_ux_messages_startup.py
#	tests/hermes_cli/test_venv_holder_classifier.py
#	tests/hermes_cli/test_verify_console_scripts.py
#	tests/hermes_cli/test_verify_core_dependencies.py
#	tests/hermes_cli/test_web_server.py
#	tests/hermes_cli/test_web_server_console_ws.py
#	tests/hermes_cli/test_web_server_ws_ping.py
#	tests/hermes_cli/test_web_ui_build.py
#	tests/hermes_state/test_fts_rebuild_admission.py
#	tests/hermes_state/test_hermes_state.py
#	tests/plugins/memory/test_memory_lazy_install.py
#	tests/plugins/test_google_meet_plugin.py
#	tests/plugins/test_langfuse_plugin.py
#	tests/plugins/test_security_guidance_plugin.py
#	tests/plugins/test_transform_llm_output_hook.py
#	tests/scripts/desktop_update/test_desktop_update_windows_gateway_flag.py
#	tests/scripts/desktop_update/test_desktop_update_windows_python_handoff.py
#	tests/scripts/desktop_update/test_desktop_update_windows_timestamp.py
#	tests/scripts/install/test_install_clone_throttle_fallback.py
#	tests/scripts/install/test_install_lockfile_churn.py
#	tests/scripts/install/test_install_no_initial_commit.py
#	tests/scripts/install/test_install_sh_browser_install.py
#	tests/scripts/install/test_install_sh_node_prerelease.py
#	tests/scripts/install/test_install_sh_symlink_stomp.py
#	tests/scripts/install/test_install_sh_uv_lock_config.py
#	tests/scripts/install/test_install_unmerged_index.py
#	tests/scripts/test_contributor_map.py
#	tests/scripts/test_run_tests_parallel.py
#	tests/skills/test_competitor_news_monitor_skill.py
#	tests/skills/test_document_to_action_items_skill.py
#	tests/skills/test_google_workspace_setup.py
#	tests/skills/test_google_workspace_setup_deps.py
#	tests/skills/test_grounded_citations_skill.py
#	tests/skills/test_ip_as_logo_skill.py
#	tests/skills/test_live_dashboard_skill.py
#	tests/skills/test_mcp_oauth_remote_gateway_skill.py
#	tests/skills/test_office_document_skills.py
#	tests/skills/test_openclaw_migration.py
#	tests/skills/test_product_price_monitor_skill.py
#	tests/skills/test_scrollcraft_skill.py
#	tests/skills/test_setup_wizard_generator_skill.py
#	tests/skills/test_weekly_review_planning_skill.py
#	tests/test_engines_satisfiable.py
#	tests/test_fast_safe_load.py
#	tests/test_hermes_bootstrap.py
#	tests/test_hermes_constants.py
#	tests/test_hermes_logging.py
#	tests/test_managed_runtime_resolution.py
#	tests/test_model_tools_async_bridge.py
#	tests/test_packaging_build_guard.py
#	tests/test_packaging_metadata.py
#	tests/test_yaml_indent_consistency.py
#	tests/tools/test_approval_timeout_overflow.py
#	tests/tools/test_base_environment.py
#	tests/tools/test_bot_mode_dm.py
#	tests/tools/test_browser_chromium_check.py
#	tests/tools/test_browser_hardening.py
#	tests/tools/test_browser_homebrew_paths.py
#	tests/tools/test_browser_npx_warmup.py
#	tests/tools/test_browser_orphan_reaper.py
#	tests/tools/test_browser_real_profile.py
#	tests/tools/test_browser_use_cli.py
#	tests/tools/test_clipboard.py
#	tests/tools/test_code_execution.py
#	tests/tools/test_code_execution_modes.py
#	tests/tools/test_code_execution_windows_env.py
#	tests/tools/test_computer_use.py
#	tests/tools/test_delegate_liveness_timeout.py
#	tests/tools/test_execute_code_approval_cluster.py
#	tests/tools/test_execution_flag_detection.py
#	tests/tools/test_fal_common.py
#	tests/tools/test_file_operations.py
#	tests/tools/test_file_tools.py
#	tests/tools/test_file_tools_cwd_resolution.py
#	tests/tools/test_file_tools_live.py
#	tests/tools/test_lazy_deps.py
#	tests/tools/test_lazy_deps_durable_target.py
#	tests/tools/test_lazy_deps_managed.py
#	tests/tools/test_local_env_blocklist.py
#	tests/tools/test_local_tempdir.py
#	tests/tools/test_macos_protected_search.py
#	tests/tools/test_mcp_npx_cached_bin.py
#	tests/tools/test_oneshot_completion_linger.py
#	tests/tools/test_process_registry.py
#	tests/tools/test_read_file_schema_gating.py
#	tests/tools/test_skill_improvements.py
#	tests/tools/test_skills_sync.py
#	tests/tools/test_termux_api_detection.py
#	tests/tools/test_tirith_security.py
#	tests/tools/test_transcription_tools.py
#	tests/tools/test_tts_streaming.py
#	tests/tools/test_wake_word.py
#	tests/tui_gateway/test_compute_host_borrowed_lease.py
#	tests/tui_gateway/test_compute_host_turn_protocol.py
#	tests/tui_gateway/test_isolated_orphan_activity.py
#	tests/tui_gateway/test_protocol.py
#	tests/tui_gateway/test_slash_worker_profile_home.py
#	tests/tui_gateway/test_subprocess_encoding.py
#	tests/tui_gateway/test_tui_gateway_server.py
#	ui-tui/src/__tests__/terminalParity.test.ts
#	ui-tui/src/__tests__/termuxComposerLayout.test.ts
#	ui-tui/src/__tests__/textInputFastEcho.test.ts
2026-09-23 07:02:44 -04:00

230 lines
8.6 KiB
Python

"""Profile-isolation regression tests for single-process multi-profile runtimes.
In runtimes that serve every profile from one OS process (the desktop
``tui_gateway``), the profile boundary is the context-local
``_HERMES_HOME_OVERRIDE`` ContextVar, not the process environment. State that
escapes the request call stack — import-time-frozen path constants, direct
``os.environ`` reads, or worker threads that don't inherit the request context —
silently reverts to the launch/default profile and leaks one profile's data
into another.
These tests drive each previously-leaking site under override A then override B
with real temp HERMES_HOME directories (no mocks) and assert the *active*
profile's path is used. They are the productionized form of the manual smoke
probes used to confirm the bug class.
"""
import os
from pathlib import Path
import pytest
from hermes_constants import (
get_hermes_home,
reset_hermes_home_override,
set_hermes_home_override,
)
@pytest.fixture
def two_profiles(tmp_path):
"""Two distinct profile HERMES_HOME dirs with the dir skeleton created."""
prof_a = tmp_path / "profA"
prof_b = tmp_path / "profB"
for p in (prof_a, prof_b):
(p / "skills").mkdir(parents=True, exist_ok=True)
(p / "state").mkdir(parents=True, exist_ok=True)
(p / "cache").mkdir(parents=True, exist_ok=True)
return prof_a, prof_b
def _under_override(home: Path, fn):
"""Run ``fn`` with the profile override set to ``home`` and reset after."""
token = set_hermes_home_override(str(home))
try:
return fn()
finally:
reset_hermes_home_override(token)
# ---------------------------------------------------------------------------
# M1 — import-time path globals / direct os.environ reads
# ---------------------------------------------------------------------------
class TestSkillsHubPathResolution:
"""tools/skills_hub.py path constants must reflect the active profile."""
def test_skills_dir_follows_override(self, two_profiles):
prof_a, prof_b = two_profiles
import tools.skills_hub as sh
# Importing/touching under A must NOT pin the path for B.
a_seen = _under_override(prof_a, lambda: Path(sh.SKILLS_DIR))
b_seen = _under_override(prof_b, lambda: Path(sh.SKILLS_DIR))
assert a_seen == prof_a / "skills"
assert b_seen == prof_b / "skills"
assert a_seen != b_seen
def test_hub_derived_paths_follow_override(self, two_profiles):
prof_a, prof_b = two_profiles
import tools.skills_hub as sh
b_lock = _under_override(prof_b, lambda: Path(sh.LOCK_FILE))
b_audit = _under_override(prof_b, lambda: Path(sh.AUDIT_LOG))
b_index = _under_override(prof_b, lambda: Path(sh.INDEX_CACHE_DIR))
assert b_lock == prof_b / "skills" / ".hub" / "lock.json"
assert b_audit == prof_b / "skills" / ".hub" / "audit.log"
assert b_index == prof_b / "skills" / ".hub" / "index-cache"
class TestGatewayCacheDirResolution:
"""gateway/platforms/base.py cache getters must follow the active profile."""
def test_image_cache_dir_follows_override(self, two_profiles):
prof_a, prof_b = two_profiles
import gateway.platforms.base as gb
a_seen = _under_override(prof_a, lambda: gb.get_image_cache_dir())
b_seen = _under_override(prof_b, lambda: gb.get_image_cache_dir())
assert str(a_seen).startswith(str(prof_a))
assert str(b_seen).startswith(str(prof_b))
assert a_seen != b_seen
class TestRichSentStorePathResolution:
"""gateway/rich_sent_store.py must honor the override, not read os.environ."""
def test_store_path_follows_override(self, two_profiles, monkeypatch):
prof_a, prof_b = two_profiles
# Ensure no ambient HERMES_HOME env masks the test.
monkeypatch.delenv("HERMES_HOME", raising=False)
import gateway.rich_sent_store as rss
b_seen = _under_override(prof_b, lambda: rss._store_path())
assert b_seen.startswith(str(prof_b))
assert b_seen.endswith(os.path.join("state", "rich_sent_index.json"))
class TestGatewayHooksDirResolution:
"""gateway/hooks.py's HookRegistry must discover hooks from the active
profile's directory — otherwise one profile's HookRegistry loads and
executes a DIFFERENT profile's hook handlers (arbitrary Python) against
its own live event context under the multiplexed gateway."""
def test_hooks_dir_follows_override(self, two_profiles):
prof_a, prof_b = two_profiles
import gateway.hooks as gh
a_seen = _under_override(prof_a, lambda: gh._resolve_hooks_dir())
b_seen = _under_override(prof_b, lambda: gh._resolve_hooks_dir())
assert a_seen == prof_a / "hooks"
assert b_seen == prof_b / "hooks"
assert a_seen != b_seen
def test_discover_and_load_uses_active_profile_hooks_dir(self, two_profiles):
"""End-to-end: a hook that only exists under profile B's hooks dir
must not be discovered when profile A's override is active, and vice
versa — proving the registry doesn't fall back to a frozen profile."""
prof_a, prof_b = two_profiles
import gateway.hooks as gh
b_hook_dir = prof_b / "hooks" / "only-in-b"
b_hook_dir.mkdir(parents=True)
(b_hook_dir / "HOOK.yaml").write_text(
"name: only-in-b\nevents: [\"agent:start\"]\n", encoding="utf-8",
)
(b_hook_dir / "handler.py").write_text(
"async def handle(event_type, context):\n pass\n", encoding="utf-8",
)
def _load_and_names():
reg = gh.HookRegistry()
reg.discover_and_load()
return [h["name"] for h in reg.loaded_hooks]
a_hooks = _under_override(prof_a, _load_and_names)
b_hooks = _under_override(prof_b, _load_and_names)
assert "only-in-b" not in a_hooks
assert "only-in-b" in b_hooks
class TestCheckpointManagerPathResolution:
"""tools/checkpoint_manager.py's checkpoint store root must honor the
active profile — otherwise one profile's CheckpointManager instance can
read/write code-edit checkpoints into a different profile's store under
the multiplexed gateway."""
def test_checkpoint_base_follows_override(self, two_profiles):
prof_a, prof_b = two_profiles
import tools.checkpoint_manager as cm
a_seen = _under_override(prof_a, lambda: cm._resolve_checkpoint_base())
b_seen = _under_override(prof_b, lambda: cm._resolve_checkpoint_base())
assert a_seen == prof_a / "checkpoints"
assert b_seen == prof_b / "checkpoints"
assert a_seen != b_seen
def test_store_path_follows_override(self, two_profiles):
prof_a, prof_b = two_profiles
import tools.checkpoint_manager as cm
b_seen = _under_override(prof_b, lambda: cm._store_path())
assert b_seen == prof_b / "checkpoints" / "store"
class TestStickerCachePathResolution:
"""gateway/sticker_cache.py's cache file must honor the active profile —
otherwise one profile's Telegram sticker-description cache leaks into a
different profile's under the multiplexed gateway."""
def test_cache_path_follows_override(self, two_profiles):
prof_a, prof_b = two_profiles
import gateway.sticker_cache as sc
a_seen = _under_override(prof_a, lambda: sc._resolve_cache_path())
b_seen = _under_override(prof_b, lambda: sc._resolve_cache_path())
assert a_seen == prof_a / "sticker_cache.json"
assert b_seen == prof_b / "sticker_cache.json"
assert a_seen != b_seen
# ---------------------------------------------------------------------------
# M2 — thread / executor context propagation
# ---------------------------------------------------------------------------
class TestThreadContextPropagation:
"""Worker threads must inherit the spawning turn's profile override."""
def test_run_async_worker_preserves_override(self, two_profiles):
"""model_tools._run_async's worker-thread branch must keep the override.
This is the generic sync->async bridge for every async tool; if it
leaks, every async tool that resolves get_hermes_home() leaks.
"""
import asyncio
_prof_a, prof_b = two_profiles
import model_tools
async def reads_home():
return str(get_hermes_home())
async def driver():
# Inside a running loop, _run_async spawns a worker thread + loop.
return model_tools._run_async(reads_home())
seen = _under_override(prof_b, lambda: asyncio.run(driver()))
assert seen == str(prof_b)