Clean up extension credentials on uninstall (#1718)

* Clean up extension credentials on uninstall

* Address PR review feedback

* Cover channel webhook secrets on uninstall

* Harden tool secret cleanup detection
This commit is contained in:
Henry Park
2026-03-28 14:46:45 +01:00
committed by GitHub
parent 8f8cb7f7b1
commit f49f368355
4 changed files with 1026 additions and 1 deletions
+2
View File
@@ -53,6 +53,7 @@ HEADED=1 pytest scenarios/
| `test_skills.py` | Skills tab UI visibility, ClawHub search (skipped if registry unreachable), install + remove lifecycle |
| `test_sse_reconnect.py` | SSE reconnects after programmatic `eventSource.close()` + `connectSSE()`; history is reloaded after reconnect |
| `test_tool_approval.py` | Approval card appears, buttons disable on approve/deny, parameters toggle via `page.evaluate("showApproval(...)")`; the waiting-approval regression uses a real HTTP tool call |
| `test_extension_uninstall_cleanup.py` | Real install/setup/remove coverage for WASM tools, WASM channels, OAuth-backed shared Google tools, and MCP servers; verifies uninstall deletes stored secrets from the libSQL `secrets` table while preserving shared credentials until the last referencing extension is removed |
| `test_oauth_refresh.py` | Hosted Gmail OAuth regression: complete setup via `/oauth/callback`, expire the stored access token in libSQL, trigger a real `gmail` tool call through `/api/chat/send`, and verify refresh goes through the mock `/oauth/refresh` proxy without forwarding `client_secret` |
## `helpers.py`
@@ -77,6 +78,7 @@ All fixtures are defined in `tests/e2e/conftest.py`. Running `pytest scenarios/`
| `mock_llm_server` | Starts `mock_llm.py --port 0`, reads the assigned port from stdout, waits for `/v1/models` to return 200. Yields the base URL. |
| `ironclaw_server` | Starts the ironclaw binary with a minimal env (see below), waits for `/api/health` (timeout 60s). Yields the base URL. On teardown sends **SIGINT** (not SIGTERM) so the tokio ctrl_c handler triggers a graceful shutdown and LLVM coverage data is flushed. |
| `hosted_oauth_refresh_server` | Starts a second ironclaw instance with a dedicated libSQL DB and `GOOGLE_OAUTH_CLIENT_ID=hosted-google-client-id`, while still pointing `IRONCLAW_OAUTH_EXCHANGE_URL` at `mock_llm.py`. Yields a dict with `base_url`, `db_path`, `gateway_user_id`, and `mock_llm_url` for the hosted refresh regression scenario. |
| `extension_cleanup_server` | Starts an isolated ironclaw instance with its own temp DB/home/WASM dirs, `SECRETS_MASTER_KEY`, and hosted-style OAuth env so uninstall-cleanup scenarios can inspect the `secrets` table without interfering with the shared E2E server state. |
| `browser` | Launches a single Chromium instance (headless by default; set `HEADED=1` for headed). Shared across all tests. |
### Function-scoped fixtures
+109
View File
@@ -443,6 +443,115 @@ async def hosted_oauth_refresh_server(
home_tmpdir.cleanup()
@pytest.fixture(scope="session")
async def extension_cleanup_server(
ironclaw_binary,
mock_llm_server,
):
"""Start an isolated ironclaw instance for uninstall secret cleanup E2E tests."""
reserved = _reserve_loopback_sockets(2)
db_tmpdir = tempfile.TemporaryDirectory(prefix="ironclaw-e2e-cleanup-db-")
home_tmpdir = tempfile.TemporaryDirectory(prefix="ironclaw-e2e-cleanup-home-")
tools_tmpdir = tempfile.TemporaryDirectory(prefix="ironclaw-e2e-cleanup-tools-")
channels_tmpdir = tempfile.TemporaryDirectory(prefix="ironclaw-e2e-cleanup-channels-")
try:
gateway_port = reserved[0].getsockname()[1]
http_port = reserved[1].getsockname()[1]
for sock in reserved:
if sock.fileno() != -1:
sock.close()
db_path = os.path.join(db_tmpdir.name, "extension-cleanup.db")
home_dir = home_tmpdir.name
env = {
"PATH": os.environ.get("PATH", "/usr/bin:/bin"),
"HOME": home_dir,
"IRONCLAW_BASE_DIR": os.path.join(home_dir, ".ironclaw"),
"RUST_LOG": "ironclaw=info",
"RUST_BACKTRACE": "1",
"IRONCLAW_OWNER_ID": OWNER_SCOPE_ID,
"GATEWAY_ENABLED": "true",
"GATEWAY_HOST": "127.0.0.1",
"GATEWAY_PORT": str(gateway_port),
"GATEWAY_AUTH_TOKEN": AUTH_TOKEN,
"GATEWAY_USER_ID": OWNER_SCOPE_ID,
"HTTP_HOST": "127.0.0.1",
"HTTP_PORT": str(http_port),
"HTTP_WEBHOOK_SECRET": HTTP_WEBHOOK_SECRET,
"CLI_ENABLED": "false",
"LLM_BACKEND": "openai_compatible",
"LLM_BASE_URL": mock_llm_server,
"LLM_MODEL": "mock-model",
"DATABASE_BACKEND": "libsql",
"LIBSQL_PATH": db_path,
"SECRETS_MASTER_KEY": "0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef",
"SANDBOX_ENABLED": "false",
"SKILLS_ENABLED": "true",
"ROUTINES_ENABLED": "true",
"HEARTBEAT_ENABLED": "false",
"EMBEDDING_ENABLED": "false",
"WASM_ENABLED": "true",
"WASM_TOOLS_DIR": tools_tmpdir.name,
"WASM_CHANNELS_DIR": channels_tmpdir.name,
"ONBOARD_COMPLETED": "true",
"IRONCLAW_OAUTH_CALLBACK_URL": "https://oauth.test.example/oauth/callback",
"IRONCLAW_OAUTH_EXCHANGE_URL": mock_llm_server,
"GOOGLE_OAUTH_CLIENT_ID": "hosted-google-client-id",
}
_forward_coverage_env(env)
proc = await asyncio.create_subprocess_exec(
ironclaw_binary, "--no-onboard",
stdin=asyncio.subprocess.DEVNULL,
stdout=asyncio.subprocess.PIPE,
stderr=asyncio.subprocess.PIPE,
env=env,
)
startup_kill_attempted = False
base_url = f"http://127.0.0.1:{gateway_port}"
try:
await wait_for_ready(f"{base_url}/api/health", timeout=60)
yield {
"base_url": base_url,
"db_path": db_path,
"gateway_user_id": OWNER_SCOPE_ID,
"mock_llm_url": mock_llm_server,
}
except TimeoutError:
if proc.returncode is None:
startup_kill_attempted = True
await _stop_process(proc, timeout=2)
returncode = proc.returncode
stderr_bytes = b""
if proc.stderr:
try:
stderr_bytes = await asyncio.wait_for(proc.stderr.read(8192), timeout=2)
except asyncio.TimeoutError:
pass
stderr_text = stderr_bytes.decode("utf-8", errors="replace")
pytest.fail(
f"extension cleanup server failed to start on port {gateway_port} "
f"(returncode={returncode}).\nstderr:\n{stderr_text}"
)
finally:
if proc.returncode is None:
if startup_kill_attempted:
await _stop_process(proc, timeout=2)
else:
await _stop_process(proc, sig=signal.SIGINT, timeout=10)
if proc.returncode is None:
await _stop_process(proc, timeout=2)
finally:
for sock in reserved:
if sock.fileno() != -1:
sock.close()
db_tmpdir.cleanup()
home_tmpdir.cleanup()
tools_tmpdir.cleanup()
channels_tmpdir.cleanup()
@pytest.fixture(scope="session")
async def http_channel_server(ironclaw_server, server_ports):
"""HTTP webhook channel base URL."""
@@ -0,0 +1,266 @@
"""Extension uninstall secret cleanup E2E tests.
Exercises real install/setup/auth/remove flows and verifies the backing
secrets table is cleaned up when extensions are uninstalled.
"""
import sqlite3
from urllib.parse import parse_qs, urlparse
import httpx
from helpers import api_get, api_post
def _extract_state(auth_url: str) -> str:
parsed = urlparse(auth_url)
state = parse_qs(parsed.query).get("state", [None])[0]
assert state, f"auth_url should include state: {auth_url}"
return state
def _secret_exists(db_path: str, user_id: str, name: str) -> bool:
with sqlite3.connect(db_path) as conn:
row = conn.execute(
"SELECT 1 FROM secrets WHERE user_id = ?1 AND name = ?2 LIMIT 1",
(user_id, name),
).fetchone()
return row is not None
def _secret_names(db_path: str, user_id: str) -> set[str]:
with sqlite3.connect(db_path) as conn:
rows = conn.execute(
"SELECT name FROM secrets WHERE user_id = ?1",
(user_id,),
).fetchall()
return {row[0] for row in rows}
async def _get_extension(base_url: str, name: str) -> dict | None:
response = await api_get(base_url, "/api/extensions", timeout=15)
response.raise_for_status()
for extension in response.json().get("extensions", []):
if extension["name"] == name:
return extension
return None
async def _ensure_removed(base_url: str, name: str) -> None:
extension = await _get_extension(base_url, name)
if extension is not None:
response = await api_post(base_url, f"/api/extensions/{name}/remove", timeout=30)
assert response.status_code == 200, response.text
assert response.json().get("success") is True, response.text
async def _install_extension(
base_url: str,
name: str,
*,
kind: str | None = None,
url: str | None = None,
) -> None:
payload = {"name": name}
if kind is not None:
payload["kind"] = kind
if url is not None:
payload["url"] = url
response = await api_post(
base_url,
"/api/extensions/install",
json=payload,
timeout=180,
)
assert response.status_code == 200, response.text
assert response.json().get("success") is True, response.text
async def test_remove_wasm_tool_deletes_unique_secret(extension_cleanup_server):
server = extension_cleanup_server["base_url"]
db_path = extension_cleanup_server["db_path"]
user_id = extension_cleanup_server["gateway_user_id"]
await _ensure_removed(server, "web-search")
await _install_extension(server, "web-search")
setup_response = await api_post(
server,
"/api/extensions/web-search/setup",
json={"secrets": {"brave_api_key": "cleanup-test-key"}},
timeout=30,
)
assert setup_response.status_code == 200, setup_response.text
assert setup_response.json().get("success") is True, setup_response.text
assert _secret_exists(db_path, user_id, "brave_api_key")
remove_response = await api_post(
server,
"/api/extensions/web-search/remove",
timeout=30,
)
assert remove_response.status_code == 200, remove_response.text
assert remove_response.json().get("success") is True, remove_response.text
assert not _secret_exists(db_path, user_id, "brave_api_key")
async def test_remove_wasm_channel_deletes_setup_secrets(extension_cleanup_server):
server = extension_cleanup_server["base_url"]
db_path = extension_cleanup_server["db_path"]
user_id = extension_cleanup_server["gateway_user_id"]
await _ensure_removed(server, "discord")
await _install_extension(server, "discord", kind="wasm_channel")
setup_response = await api_post(
server,
"/api/extensions/discord/setup",
json={
"secrets": {
"discord_bot_token": "cleanup-discord-bot-token",
"discord_public_key": "cleanup-discord-public-key",
}
},
timeout=30,
)
assert setup_response.status_code == 200, setup_response.text
assert setup_response.json().get("success") is True, setup_response.text
assert _secret_exists(db_path, user_id, "discord_bot_token")
assert _secret_exists(db_path, user_id, "discord_public_key")
remove_response = await api_post(
server,
"/api/extensions/discord/remove",
timeout=30,
)
assert remove_response.status_code == 200, remove_response.text
assert remove_response.json().get("success") is True, remove_response.text
assert not _secret_exists(db_path, user_id, "discord_bot_token")
assert not _secret_exists(db_path, user_id, "discord_public_key")
async def test_remove_shared_google_oauth_secrets_after_last_tool(extension_cleanup_server):
server = extension_cleanup_server["base_url"]
db_path = extension_cleanup_server["db_path"]
user_id = extension_cleanup_server["gateway_user_id"]
await _ensure_removed(server, "gmail")
await _ensure_removed(server, "google-drive")
await _install_extension(server, "gmail")
await _install_extension(server, "google-drive")
setup_response = await api_post(
server,
"/api/extensions/gmail/setup",
json={"secrets": {}},
timeout=30,
)
assert setup_response.status_code == 200, setup_response.text
auth_url = setup_response.json().get("auth_url")
assert auth_url, setup_response.text
async with httpx.AsyncClient() as client:
callback_response = await client.get(
f"{server}/oauth/callback",
params={"code": "mock_auth_code", "state": _extract_state(auth_url)},
timeout=30,
follow_redirects=True,
)
assert callback_response.status_code == 200, callback_response.text[:400]
shared_secrets = [
"google_oauth_token",
"google_oauth_token_refresh_token",
"google_oauth_token_scopes",
]
for secret_name in shared_secrets:
assert _secret_exists(db_path, user_id, secret_name), f"expected {secret_name} to exist"
gmail_remove_response = await api_post(
server,
"/api/extensions/gmail/remove",
timeout=30,
)
assert gmail_remove_response.status_code == 200, gmail_remove_response.text
assert gmail_remove_response.json().get("success") is True, gmail_remove_response.text
for secret_name in shared_secrets:
assert _secret_exists(db_path, user_id, secret_name), (
f"{secret_name} should remain while google-drive is still installed"
)
drive_remove_response = await api_post(
server,
"/api/extensions/google-drive/remove",
timeout=30,
)
assert drive_remove_response.status_code == 200, drive_remove_response.text
assert drive_remove_response.json().get("success") is True, drive_remove_response.text
for secret_name in shared_secrets:
assert not _secret_exists(db_path, user_id, secret_name), (
f"{secret_name} should be deleted after the last Google tool is removed"
)
async def test_remove_mcp_server_deletes_stored_secrets(extension_cleanup_server):
server = extension_cleanup_server["base_url"]
db_path = extension_cleanup_server["db_path"]
user_id = extension_cleanup_server["gateway_user_id"]
mcp_url = f"{extension_cleanup_server['mock_llm_url']}/mcp"
await _ensure_removed(server, "mock-mcp")
await _install_extension(server, "mock-mcp", kind="mcp_server", url=mcp_url)
setup_response = await api_post(
server,
"/api/extensions/mock-mcp/setup",
json={"secrets": {}},
timeout=30,
)
assert setup_response.status_code == 200, setup_response.text
auth_url = setup_response.json().get("auth_url")
if auth_url is None:
activate_response = await api_post(
server,
"/api/extensions/mock-mcp/activate",
timeout=30,
)
assert activate_response.status_code == 200, activate_response.text
auth_url = activate_response.json().get("auth_url")
assert auth_url, "mock-mcp should require OAuth in E2E"
async with httpx.AsyncClient() as client:
callback_response = await client.get(
f"{server}/oauth/callback",
params={"code": "mock_mcp_code", "state": _extract_state(auth_url)},
timeout=30,
follow_redirects=True,
)
assert callback_response.status_code == 200, callback_response.text[:400]
expected_mcp_secrets = [
"mcp_mock-mcp_access_token",
"mcp_mock-mcp_client_id",
]
stored_secret_names = _secret_names(db_path, user_id)
for secret_name in expected_mcp_secrets:
assert secret_name in stored_secret_names, (
f"expected {secret_name} to exist; stored secrets were {sorted(stored_secret_names)}"
)
remove_response = await api_post(
server,
"/api/extensions/mock-mcp/remove",
timeout=30,
)
assert remove_response.status_code == 200, remove_response.text
assert remove_response.json().get("success") is True, remove_response.text
remaining_secret_names = _secret_names(db_path, user_id)
assert not any(name.startswith("mcp_mock-mcp_") for name in remaining_secret_names), (
f"mock-mcp secrets should be deleted on remove; remaining secrets were "
f"{sorted(remaining_secret_names)}"
)