hermes-agent/tests/run_agent/test_agent_guardrails.py
Teknium 1a47769715
test: deflake CI and dev-machine flaky tests in bulk (11 tests, 10 files) (#61816)
* test: deflake CI and dev-machine flaky tests in bulk

Fixes ten distinct flake sources found by mining recent CI failures and
running the full suite on a dev machine with real user state:

CI-observed races:
- tests/conftest.py live-system guard: allow signal 0 (pure liveness
  probe) through _guarded_kill/_guarded_killpg. psutil.pid_exists()
  probes a just-killed grandchild reparented to init; the subtree check
  fails for it and the guard RuntimeError'd
  test_entire_tree_is_sigkilled_not_just_parent intermittently on
  unrelated PRs.

Hermeticity flakes (fail on dev machines with real state, pass on CI):
- agent/coding_context.py: _marker_root() now skips the shared temp
  root (tempfile.gettempdir()) like it skips $HOME — a stray
  /tmp/package.json flipped every tmp_path test into the coding
  posture (9 failures in test_coding_context.py).
- test_agent_guardrails.py: pin MAX_CONCURRENT_CHILDREN=3 via autouse
  monkeypatch instead of freezing the user's real config value at
  import time (import-time vs call-time config mismatch).
- test_web_tools_config.py: TestCheckWebApiKey now neutralizes the
  ddgs package probe and registry providers — the optional ddgs
  package in a dev venv lit up the fallback backend.
- test_credential_pool.py: block claude_code/hermes-oauth credential
  autodiscovery in the two pool-merge tests that assert exact id
  lists (a real ~/.claude/.credentials.json seeded an extra entry).
- test_modal_sandbox_fixes.py: clear _permanent_approved /
  _session_approved — the user's real command_allowlist silently
  approved the guard-escalation commands under test.
- test_setup_irc.py: stub prompt_checklist to select only the IRC row;
  the non-TTY cancel fallback re-ran the real configured platforms'
  interactive setup_fn, which hit input() under captured stdin.
- test_doctor.py: TestGitHubTokenCheck now patches the module-level
  HERMES_HOME constant (the file's established pattern) instead of
  only setenv — doctor was running PRAGMA integrity_check against the
  real multi-GB state.db and blowing the 300s per-file budget.

Latent atexit-duplication (same _enter_buffered_busy class as #34217):
- test_undo_command.py: drop importlib.reload(tui_gateway.server) in
  fixture teardown; reload re-registers the module's atexit hooks.
- test_session_platform_resolution.py: drop per-test reload of
  tui_gateway.server; every resolver reads env at call time.

* test: sentinel model value in ignore-user-config fallback assertion

With HERMES_IGNORE_USER_CONFIG=1, load_cli_config() falls back to the
repo-root cli-config.yaml (untracked, gitignored). On a dev machine that
file can legitimately set the same popular model the test hardcoded
(anthropic/claude-sonnet-4.6), flipping the != assertion locally while
CI (no cli-config.yaml) stayed green. Use an impossible sentinel model
name instead.
2026-07-09 20:03:11 -07:00

336 lines
12 KiB
Python

"""Unit tests for AIAgent pre/post-LLM-call guardrails.
Covers three static methods on AIAgent (inspired by PR #1321 — @alireza78a):
- _sanitize_api_messages() — Phase 1: orphaned tool pair repair
- _cap_delegate_task_calls() — Phase 2a: subagent concurrency limit
- _deduplicate_tool_calls() — Phase 2b: identical call deduplication
"""
import types
import pytest
from run_agent import AIAgent
# Pin the concurrency limit instead of reading the runtime config.
# _cap_delegate_task_calls() resolves _get_max_concurrent_children() at CALL
# time (inside a per-test hermetic HERMES_HOME), but this module previously
# froze the value at IMPORT time — before the hermetic fixture ran — so a
# developer machine with delegation.max_concurrent_children in the real
# ~/.hermes/config.yaml saw a different limit at import vs call and the
# truncation tests failed locally while passing on CI.
MAX_CONCURRENT_CHILDREN = 3
@pytest.fixture(autouse=True)
def _pin_max_concurrent_children(monkeypatch):
monkeypatch.setattr(
"tools.delegate_tool._get_max_concurrent_children",
lambda: MAX_CONCURRENT_CHILDREN,
)
# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------
def make_tc(name: str, arguments: str = "{}") -> types.SimpleNamespace:
"""Create a minimal tool_call SimpleNamespace mirroring the OpenAI SDK object."""
tc = types.SimpleNamespace()
tc.function = types.SimpleNamespace(name=name, arguments=arguments)
return tc
def tool_result(call_id: str, content: str = "ok") -> dict:
return {"role": "tool", "tool_call_id": call_id, "content": content}
def assistant_dict_call(call_id: str, name: str = "terminal") -> dict:
"""Dict-style tool_call (as stored in message history)."""
return {"id": call_id, "function": {"name": name, "arguments": "{}"}}
# ---------------------------------------------------------------------------
# Phase 1 — _sanitize_api_messages
# ---------------------------------------------------------------------------
class TestSanitizeApiMessages:
def test_orphaned_result_removed(self):
msgs = [
{"role": "assistant", "tool_calls": [assistant_dict_call("c1")]},
tool_result("c1"),
tool_result("c_ORPHAN"),
]
out = AIAgent._sanitize_api_messages(msgs)
assert len(out) == 2
assert all(m.get("tool_call_id") != "c_ORPHAN" for m in out)
def test_orphaned_call_gets_stub_result(self):
msgs = [
{"role": "assistant", "tool_calls": [assistant_dict_call("c2")]},
]
out = AIAgent._sanitize_api_messages(msgs)
assert len(out) == 2
stub = out[1]
assert stub["role"] == "tool"
assert stub["tool_call_id"] == "c2"
assert stub["content"]
def test_clean_messages_pass_through(self):
msgs = [
{"role": "user", "content": "hello"},
{"role": "assistant", "tool_calls": [assistant_dict_call("c3")]},
tool_result("c3"),
{"role": "assistant", "content": "done"},
]
out = AIAgent._sanitize_api_messages(msgs)
assert out == msgs
def test_mixed_orphaned_result_and_orphaned_call(self):
msgs = [
{"role": "assistant", "tool_calls": [
assistant_dict_call("c4"),
assistant_dict_call("c5"),
]},
tool_result("c4"),
tool_result("c_DANGLING"),
]
out = AIAgent._sanitize_api_messages(msgs)
ids = [m.get("tool_call_id") for m in out if m.get("role") == "tool"]
assert "c_DANGLING" not in ids
assert "c4" in ids
assert "c5" in ids
def test_empty_list_is_safe(self):
assert AIAgent._sanitize_api_messages([]) == []
def test_no_tool_messages(self):
msgs = [
{"role": "user", "content": "hi"},
{"role": "assistant", "content": "hello"},
]
out = AIAgent._sanitize_api_messages(msgs)
assert out == msgs
def test_sdk_object_tool_calls(self):
tc_obj = types.SimpleNamespace(id="c6", function=types.SimpleNamespace(
name="terminal", arguments="{}"
))
msgs = [
{"role": "assistant", "tool_calls": [tc_obj]},
]
out = AIAgent._sanitize_api_messages(msgs)
assert len(out) == 2
assert out[1]["tool_call_id"] == "c6"
def test_tool_result_with_leading_whitespace_preserved(self):
"""Tool result IDs with leading/trailing whitespace should match assistant call IDs."""
msgs = [
{"role": "assistant", "tool_calls": [assistant_dict_call("functions.cronjob:24")]},
tool_result(" functions.cronjob:24"), # leading whitespace
]
out = AIAgent._sanitize_api_messages(msgs)
# Should NOT inject a stub — the tool result is valid after stripping
assert len(out) == 2
assert out[1]["role"] == "tool"
assert out[1]["content"] == "ok"
def test_truly_orphaned_with_whitespace_still_removed(self):
"""Truly orphaned tool results with whitespace should still be removed."""
msgs = [
{"role": "assistant", "tool_calls": [assistant_dict_call("c_valid")]},
tool_result(" c_ORPHAN "), # whitespace + no matching call
]
out = AIAgent._sanitize_api_messages(msgs)
assert len(out) == 2 # assistant + stub for c_valid, orphan removed
tool_msgs = [m for m in out if m["role"] == "tool"]
assert len(tool_msgs) == 1
assert tool_msgs[0]["tool_call_id"] == "c_valid"
# ---------------------------------------------------------------------------
# Phase 2a — _cap_delegate_task_calls
# ---------------------------------------------------------------------------
class TestCapDelegateTaskCalls:
def test_excess_delegates_truncated(self):
tcs = [make_tc("delegate_task") for _ in range(MAX_CONCURRENT_CHILDREN + 2)]
out = AIAgent._cap_delegate_task_calls(tcs)
delegate_count = sum(1 for tc in out if tc.function.name == "delegate_task")
assert delegate_count == MAX_CONCURRENT_CHILDREN
def test_non_delegate_calls_preserved(self):
tcs = (
[make_tc("delegate_task") for _ in range(MAX_CONCURRENT_CHILDREN + 1)]
+ [make_tc("terminal"), make_tc("web_search")]
)
out = AIAgent._cap_delegate_task_calls(tcs)
names = [tc.function.name for tc in out]
assert "terminal" in names
assert "web_search" in names
def test_at_limit_passes_through(self):
tcs = [make_tc("delegate_task") for _ in range(MAX_CONCURRENT_CHILDREN)]
out = AIAgent._cap_delegate_task_calls(tcs)
assert out is tcs
def test_below_limit_passes_through(self):
tcs = [make_tc("delegate_task") for _ in range(MAX_CONCURRENT_CHILDREN - 1)]
out = AIAgent._cap_delegate_task_calls(tcs)
assert out is tcs
def test_no_delegate_calls_unchanged(self):
tcs = [make_tc("terminal"), make_tc("web_search")]
out = AIAgent._cap_delegate_task_calls(tcs)
assert out is tcs
def test_empty_list_safe(self):
assert AIAgent._cap_delegate_task_calls([]) == []
def test_original_list_not_mutated(self):
tcs = [make_tc("delegate_task") for _ in range(MAX_CONCURRENT_CHILDREN + 2)]
original_len = len(tcs)
AIAgent._cap_delegate_task_calls(tcs)
assert len(tcs) == original_len
def test_interleaved_order_preserved(self):
delegates = [make_tc("delegate_task", f'{{"task":"{i}"}}')
for i in range(MAX_CONCURRENT_CHILDREN + 1)]
t1 = make_tc("terminal", '{"cmd":"ls"}')
w1 = make_tc("web_search", '{"q":"x"}')
tcs = [delegates[0], t1, delegates[1], w1] + delegates[2:]
out = AIAgent._cap_delegate_task_calls(tcs)
expected = [delegates[0], t1, delegates[1], w1] + delegates[2:MAX_CONCURRENT_CHILDREN]
assert len(out) == len(expected)
for i, (actual, exp) in enumerate(zip(out, expected)):
assert actual is exp, f"mismatch at index {i}"
# ---------------------------------------------------------------------------
# Phase 2b — _deduplicate_tool_calls
# ---------------------------------------------------------------------------
class TestDeduplicateToolCalls:
def test_duplicate_pair_deduplicated(self):
tcs = [
make_tc("web_search", '{"query":"foo"}'),
make_tc("web_search", '{"query":"foo"}'),
]
out = AIAgent._deduplicate_tool_calls(tcs)
assert len(out) == 1
def test_multiple_duplicates(self):
tcs = [
make_tc("web_search", '{"q":"a"}'),
make_tc("web_search", '{"q":"a"}'),
make_tc("terminal", '{"cmd":"ls"}'),
make_tc("terminal", '{"cmd":"ls"}'),
make_tc("terminal", '{"cmd":"pwd"}'),
]
out = AIAgent._deduplicate_tool_calls(tcs)
assert len(out) == 3
def test_same_tool_different_args_kept(self):
tcs = [
make_tc("terminal", '{"cmd":"ls"}'),
make_tc("terminal", '{"cmd":"pwd"}'),
]
out = AIAgent._deduplicate_tool_calls(tcs)
assert out is tcs
def test_different_tools_same_args_kept(self):
tcs = [
make_tc("tool_a", '{"x":1}'),
make_tc("tool_b", '{"x":1}'),
]
out = AIAgent._deduplicate_tool_calls(tcs)
assert out is tcs
def test_clean_list_unchanged(self):
tcs = [
make_tc("web_search", '{"q":"x"}'),
make_tc("terminal", '{"cmd":"ls"}'),
]
out = AIAgent._deduplicate_tool_calls(tcs)
assert out is tcs
def test_empty_list_safe(self):
assert AIAgent._deduplicate_tool_calls([]) == []
def test_first_occurrence_kept(self):
tc1 = make_tc("terminal", '{"cmd":"ls"}')
tc2 = make_tc("terminal", '{"cmd":"ls"}')
out = AIAgent._deduplicate_tool_calls([tc1, tc2])
assert len(out) == 1
assert out[0] is tc1
def test_original_list_not_mutated(self):
tcs = [
make_tc("web_search", '{"q":"dup"}'),
make_tc("web_search", '{"q":"dup"}'),
]
original_len = len(tcs)
AIAgent._deduplicate_tool_calls(tcs)
assert len(tcs) == original_len
# ---------------------------------------------------------------------------
# _get_tool_call_id_static
# ---------------------------------------------------------------------------
class TestGetToolCallIdStatic:
def test_dict_with_valid_id(self):
assert AIAgent._get_tool_call_id_static({"id": "call_123"}) == "call_123"
def test_dict_with_none_id(self):
assert AIAgent._get_tool_call_id_static({"id": None}) == ""
def test_dict_without_id_key(self):
assert AIAgent._get_tool_call_id_static({"function": {}}) == ""
def test_object_with_valid_id(self):
tc = types.SimpleNamespace(id="call_456")
assert AIAgent._get_tool_call_id_static(tc) == "call_456"
def test_object_with_none_id(self):
tc = types.SimpleNamespace(id=None)
assert AIAgent._get_tool_call_id_static(tc) == ""
def test_object_without_id_attr(self):
tc = types.SimpleNamespace()
assert AIAgent._get_tool_call_id_static(tc) == ""
# ---------------------------------------------------------------------------
# _get_tool_call_name_static
# ---------------------------------------------------------------------------
class TestGetToolCallNameStatic:
def test_dict_with_valid_name(self):
assert AIAgent._get_tool_call_name_static(
{"id": "call_1", "function": {"name": "terminal", "arguments": "{}"}}
) == "terminal"
def test_dict_with_missing_function(self):
assert AIAgent._get_tool_call_name_static({"id": "call_1"}) == ""
def test_dict_with_none_function(self):
assert AIAgent._get_tool_call_name_static({"id": "call_1", "function": None}) == ""
def test_dict_with_none_name(self):
assert AIAgent._get_tool_call_name_static(
{"function": {"name": None, "arguments": "{}"}}
) == ""
def test_object_with_valid_name(self):
tc = make_tc("read_file")
assert AIAgent._get_tool_call_name_static(tc) == "read_file"
def test_object_without_function_attr(self):
tc = types.SimpleNamespace(id="call_1")
assert AIAgent._get_tool_call_name_static(tc) == ""