mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-07-31 19:16:29 +00:00
6b81590c55
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).
136 lines
5.2 KiB
Python
136 lines
5.2 KiB
Python
"""Tests for the thinking-only assistant message sanitizer.
|
|
|
|
Covers _is_thinking_only_assistant() + _drop_thinking_only_and_merge_users()
|
|
in run_agent.py. The sanitizer runs on the per-call api_messages copy and
|
|
drops assistant turns that contain only reasoning (no visible content, no
|
|
tool_calls). Adjacent user messages left behind are merged so role
|
|
alternation is preserved for the provider.
|
|
|
|
Claude Code uses this exact pattern (filterOrphanedThinkingOnlyMessages +
|
|
mergeAdjacentUserMessages in src/utils/messages.ts). See #16823 for the
|
|
backstory on why the alternative — fabricating "." stub text — was rejected.
|
|
"""
|
|
|
|
from run_agent import AIAgent
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# _is_thinking_only_assistant — detection
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
class TestIsThinkingOnlyAssistant:
|
|
|
|
def test_plain_assistant_reply_is_not_thinking_only(self):
|
|
msg = {"role": "assistant", "content": "Hello there"}
|
|
assert not AIAgent._is_thinking_only_assistant(msg)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_non_dict_returns_false(self):
|
|
assert not AIAgent._is_thinking_only_assistant(None)
|
|
assert not AIAgent._is_thinking_only_assistant("hello")
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# _drop_thinking_only_and_merge_users — the full pass
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
class TestDropThinkingOnlyAndMergeUsers:
|
|
|
|
def test_empty_list_passthrough(self):
|
|
assert AIAgent._drop_thinking_only_and_merge_users([]) == []
|
|
|
|
def test_no_thinking_only_messages_is_noop_identity(self):
|
|
msgs = [
|
|
{"role": "user", "content": "hi"},
|
|
{"role": "assistant", "content": "hello"},
|
|
]
|
|
out = AIAgent._drop_thinking_only_and_merge_users(msgs)
|
|
# Should return the original list untouched (identity) when no changes.
|
|
assert out is msgs
|
|
|
|
|
|
def test_preserves_alternation_after_drop(self):
|
|
msgs = [
|
|
{"role": "user", "content": "u1"},
|
|
{"role": "assistant", "content": "", "reasoning": "..."},
|
|
{"role": "user", "content": "u2"},
|
|
{"role": "assistant", "content": "real reply"},
|
|
]
|
|
out = AIAgent._drop_thinking_only_and_merge_users(msgs)
|
|
roles = [m["role"] for m in out]
|
|
assert roles == ["user", "assistant"]
|
|
assert out[0]["content"] == "u1\n\nu2"
|
|
assert out[1]["content"] == "real reply"
|
|
|
|
|
|
|
|
def test_does_not_touch_stored_messages_original_list_unmutated(self):
|
|
original_first_user = {"role": "user", "content": "u1"}
|
|
original_assistant = {"role": "assistant", "content": "", "reasoning": "..."}
|
|
original_second_user = {"role": "user", "content": "u2"}
|
|
msgs = [original_first_user, original_assistant, original_second_user]
|
|
AIAgent._drop_thinking_only_and_merge_users(msgs)
|
|
# Caller passes in a per-call copy already, but the sanitizer itself
|
|
# must not rewrite the dicts it was handed on the drop path.
|
|
# (It CAN mutate merged dicts — those come from the caller's copy.)
|
|
assert original_first_user["content"] == "u1"
|
|
assert original_second_user["content"] == "u2"
|
|
|
|
def test_tool_result_between_user_and_thinking_preserved(self):
|
|
# Tool results shouldn't block a drop — but they do block the merge
|
|
# (user/tool are different roles). This scenario shouldn't happen in
|
|
# practice because a thinking-only turn won't have tool_calls, but if
|
|
# it did somehow, the surrounding tool result stays put.
|
|
msgs = [
|
|
{"role": "user", "content": "u1"},
|
|
{"role": "assistant", "tool_calls": [{"id": "c1", "function": {"name": "t", "arguments": "{}"}}]},
|
|
{"role": "tool", "tool_call_id": "c1", "content": "ok"},
|
|
{"role": "assistant", "content": "", "reasoning": "..."},
|
|
{"role": "user", "content": "u2"},
|
|
]
|
|
out = AIAgent._drop_thinking_only_and_merge_users(msgs)
|
|
assert [m["role"] for m in out] == ["user", "assistant", "tool", "user"]
|
|
|
|
def test_merge_concatenates_list_content_user_messages(self):
|
|
msgs = [
|
|
{"role": "user", "content": [{"type": "text", "text": "first"}]},
|
|
{"role": "assistant", "content": "", "reasoning": "..."},
|
|
{"role": "user", "content": [{"type": "text", "text": "second"}]},
|
|
]
|
|
out = AIAgent._drop_thinking_only_and_merge_users(msgs)
|
|
assert len(out) == 1
|
|
assert out[0]["content"] == [
|
|
{"type": "text", "text": "first"},
|
|
{"type": "text", "text": "second"},
|
|
]
|
|
|
|
|
|
def test_system_messages_ignored_by_pass(self):
|
|
msgs = [
|
|
{"role": "system", "content": "sys prompt"},
|
|
{"role": "user", "content": "u1"},
|
|
{"role": "assistant", "content": "", "reasoning": "..."},
|
|
{"role": "user", "content": "u2"},
|
|
]
|
|
out = AIAgent._drop_thinking_only_and_merge_users(msgs)
|
|
assert len(out) == 2
|
|
assert out[0]["role"] == "system"
|
|
assert out[1]["role"] == "user"
|
|
assert out[1]["content"] == "u1\n\nu2"
|