hermes-agent/tests/agent/test_tool_call_arg_no_redaction.py
Teknium 6b81590c55
test: prune low-value tests suite-wide (wave 1) — 46,820 → 28,106 test functions
Systematic prune per AGENTS.md test policy, one pass over every major
test tree (gateway, hermes_cli, tools, agent, run_agent, plugins, cli,
cron, tui_gateway, honcho/openviking, root-level):

- DELETE: source-reading tests (read_text/getsource on prod files),
  change-detector tests (exact catalog counts, model-name snapshots,
  config version literals), mock-echo tests (assert a mock returns what
  it was told), assertion-free/trivial tests, near-duplicate
  parametrizations (boundaries + one representative kept), async/sync
  twin duplicates, cosmetic within-file variations.
- KEEP (mandatory): security/redaction/approval guards, message-role
  alternation invariants, prompt-caching/deterministic-call-id
  invariants, issue-number regression tests (deduped), E2E tests.
- 6 test files deleted outright (script-style/no-assert or fully
  redundant); conftest.py, fakes/, fixtures/ untouched.
- tests/acp/conftest.py added: autouse fixture stubs the live
  models.dev/GitHub/Copilot/Anthropic inventory fetches that ACP server
  tests performed on every session create — test_server.py 147s → 3.4s,
  and the tests are now genuinely hermetic.
- Sleep-based slowness shrunk where safe (codex_ttfb_watchdog,
  compression_concurrent_fork, etc.); no wall-clock assertion tightened.

Verification: full hermetic suite via scripts/run_tests.sh —
2439 files, 31,130 tests passed, 0 failed, 0 flaky retries, 315s wall
(baseline: 583s wall, 13,564s subprocess CPU).
2026-07-29 13:10:23 -07:00

84 lines
2.5 KiB
Python

"""Regression test for #43083.
``build_assistant_message`` must NOT redact tool-call arguments. The dict it
returns enters the in-memory conversation history that is replayed to the model
on every subsequent turn AND is persisted to state.db, which is itself replayed
verbatim on session resume. Masking a credential to ``***`` there poisons the
replay: the model reads back its own ``PGPASSWORD='***' psql ...`` call and
copies the placeholder into the next tool call, breaking every
credential-dependent command on the second turn.
"""
from unittest.mock import MagicMock
from agent.chat_completion_helpers import build_assistant_message
class _FakeToolCall:
def __init__(self, tc_id, name, arguments):
self.id = tc_id
self.type = "function"
self.function = MagicMock()
self.function.name = name
self.function.arguments = arguments
self.extra_content = None
def __getattr__(self, _name):
return None
class _FakeAssistantMsg:
def __init__(self, content, tool_calls):
self.content = content
self.tool_calls = tool_calls
self.function_call = None
self.reasoning_content = None
self.model_extra = None
self.reasoning_details = None
def __getattr__(self, _name):
return None
class _FakeAgent:
stream_delta_callback = None
_stream_callback = None
reasoning_callback = None
verbose_logging = False
def _extract_reasoning(self, _msg):
return None
def _strip_think_blocks(self, text):
return text
def _needs_thinking_reasoning_pad(self):
return False
def _split_responses_tool_id(self, _raw):
return (None, None)
def _derive_responses_function_call_id(self, _call_id, _resp_id):
return None
def _deterministic_call_id(self, _name, _args, idx):
return f"det_{idx}"
def _build(arguments):
tc = _FakeToolCall("call_1", "terminal", arguments)
msg = build_assistant_message(_FakeAgent(), _FakeAssistantMsg("ok", [tc]), "tool_calls")
return msg["tool_calls"][0]["function"]["arguments"]
def test_pgpassword_preserved_verbatim(monkeypatch):
# Force redaction ON to prove build_assistant_message bypasses it for
# tool-call args regardless of the global toggle.
monkeypatch.setattr("agent.redact._REDACT_ENABLED", True, raising=False)
args = '{"command": "PGPASSWORD=\'honchorulez\' psql -h 127.0.0.1"}'
got = _build(args)
assert got == args
assert "honchorulez" in got
assert "***" not in got