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).
127 lines
3.7 KiB
Python
127 lines
3.7 KiB
Python
"""Tests for AIAgent._repair_tool_call — tool-name normalization.
|
|
|
|
Regression guard for #14784: Claude-style models sometimes emit
|
|
class-like tool-call names (``TodoTool_tool``, ``Patch_tool``,
|
|
``BrowserClick_tool``, ``PatchTool``). Before the fix they returned
|
|
"Unknown tool" even though the target tool was registered under a
|
|
snake_case name. The repair routine now normalizes CamelCase,
|
|
strips trailing ``_tool`` / ``-tool`` / ``tool`` suffixes (up to
|
|
twice to handle double-tacked suffixes like ``TodoTool_tool``), and
|
|
falls back to fuzzy match.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
from types import SimpleNamespace
|
|
|
|
import pytest
|
|
|
|
|
|
VALID = {
|
|
"todo",
|
|
"patch",
|
|
"browser_click",
|
|
"browser_navigate",
|
|
"web_search",
|
|
"read_file",
|
|
"write_file",
|
|
"terminal",
|
|
"execute_code",
|
|
"session_search",
|
|
}
|
|
|
|
|
|
@pytest.fixture
|
|
def repair():
|
|
"""Return a bound _repair_tool_call built on a minimal shell agent.
|
|
|
|
We avoid constructing a real AIAgent (which pulls in credential
|
|
resolution, session DB, etc.) because the repair routine only
|
|
reads self.valid_tool_names. A SimpleNamespace stub is enough to
|
|
bind the unbound function.
|
|
"""
|
|
from run_agent import AIAgent
|
|
stub = SimpleNamespace(valid_tool_names=VALID)
|
|
return AIAgent._repair_tool_call.__get__(stub, AIAgent)
|
|
|
|
|
|
class TestExistingBehaviorStillWorks:
|
|
"""Pre-existing repairs must keep working (no regressions)."""
|
|
|
|
def test_lowercase_already_matches(self, repair):
|
|
assert repair("browser_click") == "browser_click"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class TestClassLikeEmissions:
|
|
"""Regression coverage for #14784 — CamelCase + _tool suffix variants."""
|
|
|
|
def test_camel_case_no_suffix(self, repair):
|
|
assert repair("BrowserClick") == "browser_click"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class TestEdgeCases:
|
|
"""Edge inputs that must not crash or produce surprising results."""
|
|
|
|
def test_empty_string(self, repair):
|
|
assert repair("") is None
|
|
|
|
|
|
|
|
|
|
|
|
class TestVolcEngineXmlPollution:
|
|
"""Regression coverage for #33007 — VolcEngine ``api/plan`` endpoint
|
|
leaks raw XML attribute fragments into ``tool_use.name``.
|
|
|
|
Observed in production with the ``anthropic_messages`` API mode:
|
|
|
|
terminal" parameter="command" string="true
|
|
execute_code" parameter="code" string="true
|
|
session_search" parameter="session_id" string="true
|
|
|
|
The fix trims at the first ``"``/``'``/``<``/``>`` so the rest of
|
|
the repair pipeline can resolve the cleaned name to a real tool.
|
|
"""
|
|
|
|
def test_terminal_with_xml_attribute_pollution(self, repair):
|
|
# Exact pattern from the bug report (terminal call).
|
|
polluted = 'terminal" parameter="command" string="true'
|
|
assert repair(polluted) == "terminal"
|
|
|
|
|
|
|
|
|
|
def test_tool_name_with_trailing_quote_only(self, repair):
|
|
# Minimal leak — just a stray trailing quote, no full attribute.
|
|
assert repair('terminal"') == "terminal"
|
|
|
|
|
|
|
|
def test_clean_tool_name_unaffected_by_sanitizer(self, repair):
|
|
# Pure passthrough — no XML/quote chars, no change.
|
|
assert repair("execute_code") == "execute_code"
|
|
assert repair("session_search") == "session_search"
|
|
|
|
def test_space_separated_name_still_normalizes(self, repair):
|
|
# Critical: the XML strip must NOT consume whitespace, or the
|
|
# legitimate ``"write file" -> write_file`` repair path breaks.
|
|
assert repair("write file") == "write_file"
|
|
|
|
|
|
def test_leading_quote_falls_through_to_fuzzy_match(self, repair):
|
|
# Sanitizer only trims when the XML char is at idx > 0 — a
|
|
# name that *starts* with a quote is left untouched so the
|
|
# rest of the pipeline (fuzzy match at 0.7 cutoff) can still
|
|
# recover the obvious target.
|
|
assert repair('"terminal"') == "terminal"
|