The agent that grows with you
Python aiai-agentai-agentsanthropicchatgptclaudeclaude-codecodexhermeshermes-agentllmnous-researchopenai
"""``VaultStore._read_all`` must surface every decryptable-but-malformed vault
file as ``VaultError`` so callers that catch ``VaultError`` (the ``hermes vault``
commands, ``vault.*`` RPC handlers) get the designed error path instead of a raw
traceback. Corruption arrives via partial-write survivors, manual edits of the
envelope, or format drift."""
import pytest
from agent.vault_store import VaultError, VaultStore
def _corrupt(store: VaultStore, payload: bytes) -> None:
store._vault_path.write_bytes(store._fernet().encrypt(payload))
def _store(tmp_path) -> VaultStore:
store = VaultStore(tmp_path / "vault")
store.add_item(
"login",
"Example login",
{"identifier_type": "email", "identifier": "[email protected]", "password": "pw"},
origin="https://example.com",
)
return store
@pytest.mark.parametrize(
"op",
["list_items", "get_meta", "remove_item", "resolve_secret", "add_item"],
)
def test_corrupt_vault_raises_vault_error_on_every_reader(tmp_path, op):
"""Every ``_read_all`` consumer must see the same VaultError contract so
callers that catch VaultError all get the designed error path."""
store = _store(tmp_path)
_corrupt(store, b"{not json")
call = {
"list_items": lambda: store.list_items(),
"get_meta": lambda: store.get_meta("vault_x"),
"remove_item": lambda: store.remove_item("vault_x"),
"resolve_secret": lambda: store.resolve_secret("vault_x"),
"add_item": lambda: store.add_item(
"login",
"x",
{"identifier_type": "email", "identifier": "u", "password": "p"},
origin="https://x.example",
),
}[op]
with pytest.raises(VaultError):
call()