diff --git a/AGENTS.md b/AGENTS.md
index d17fe4cb..bb53cc0a 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -79,7 +79,7 @@ existing async and typing patterns unless the task explicitly requires a new con
 
 - CLI syntax is `reme ACTION key=value ...`; leading `-` or `--` on arguments is accepted.
 - Nested overrides use dot notation. Values support null, booleans, numbers, JSON collections, and quoted JSON strings;
-  leading-zero numeric-looking values remain strings.
+  leading-zero numeric-looking values, `nan`, and `inf` remain strings.
 - `config=<name-or-path>` loads a discovered config name or a `.yaml`, `.yml`, or `.json` file. With no explicit config
   path, `default` is loaded when available.
 - Config files expand `${VAR}` and `${VAR:-default}` recursively. An undefined variable without a default is an error.
diff --git a/docs/en/configuration.md b/docs/en/configuration.md
index 002d2b11..304d140c 100644
--- a/docs/en/configuration.md
+++ b/docs/en/configuration.md
@@ -42,7 +42,7 @@ Arguments use `key=value`; leading `-` or `--` is accepted:
 reme start --service.port=8181 --service.web_enabled=false
 ```
 
-Values support null, booleans, numbers, JSON arrays and objects, quoted JSON strings, and plain strings. Numeric-looking values with leading zeroes, such as `007`, remain strings. Quote values such as `"true"` in JSON when they must remain strings.
+Values support null, booleans, numbers, JSON arrays and objects, quoted JSON strings, and plain strings. Numeric-looking values with leading zeroes, such as `007`, remain strings, as do `nan`, `inf`, and `Infinity`. Quote values such as `"true"` in JSON when they must remain strings.
 
 ## Environment variables
 
diff --git a/docs/zh/configuration.md b/docs/zh/configuration.md
index dd046f4e..818f4f4e 100644
--- a/docs/zh/configuration.md
+++ b/docs/zh/configuration.md
@@ -47,7 +47,7 @@ reme start --service.port=8181 --service.web_enabled=false
 - JSON 引号字符串；
 - 普通字符串。
 
-类似 `007` 的前导零字符串不会被转换成数字。需要保留 `true`、`false` 等字面字符串时，使用 JSON 引号：`value='"true"'`。
+类似 `007` 的前导零字符串以及 `nan`、`inf`、`Infinity` 不会被转换成数字。需要保留 `true`、`false` 等字面字符串时，使用 JSON 引号：`value='"true"'`。
 
 ## 环境变量
 
diff --git a/plugins/beam/src/reme_beam/search_v2.py b/plugins/beam/src/reme_beam/search_v2.py
index 77c02027..aa606a69 100644
--- a/plugins/beam/src/reme_beam/search_v2.py
+++ b/plugins/beam/src/reme_beam/search_v2.py
@@ -99,7 +99,8 @@ class SearchV2Step(_ToolContextDedupMixin, BaseStep):
 
     async def execute(self):
         assert self.context is not None
-        query: str = (self.context.get("query", "") or "").strip()
+        raw_query = self.context.get("query")
+        query: str = "" if raw_query is None else str(raw_query).strip()
         limit: int = int(self.context.get("limit") or _default_limit())
         min_score: float = float(self.context.get("min_score") or 0.0)
         # vector_weight: prefer agent-supplied context value; fallback to YAML kwargs / default 0.7.
diff --git a/plugins/lme/src/reme_lme/search_v2.py b/plugins/lme/src/reme_lme/search_v2.py
index 77c02027..aa606a69 100644
--- a/plugins/lme/src/reme_lme/search_v2.py
+++ b/plugins/lme/src/reme_lme/search_v2.py
@@ -99,7 +99,8 @@ class SearchV2Step(_ToolContextDedupMixin, BaseStep):
 
     async def execute(self):
         assert self.context is not None
-        query: str = (self.context.get("query", "") or "").strip()
+        raw_query = self.context.get("query")
+        query: str = "" if raw_query is None else str(raw_query).strip()
         limit: int = int(self.context.get("limit") or _default_limit())
         min_score: float = float(self.context.get("min_score") or 0.0)
         # vector_weight: prefer agent-supplied context value; fallback to YAML kwargs / default 0.7.
diff --git a/reme/config/config_parser.py b/reme/config/config_parser.py
index ee06edd5..64db3f88 100644
--- a/reme/config/config_parser.py
+++ b/reme/config/config_parser.py
@@ -1,6 +1,7 @@
 """Parser for YAML config with CLI argument overrides."""
 
 import json
+import math
 import os
 import re
 from collections.abc import Mapping
@@ -101,6 +102,7 @@ def _convert_value(value_str: str) -> Any:
     Only converts "true"/"false" (case-insensitive) to boolean.
     Use JSON format (e.g., '"yes"', '"no"') to preserve these as strings.
     Leading-zero strings (e.g., "007", "00501") are kept as strings.
+    "nan", "inf" and "Infinity" are kept as strings: JSON has no such numbers.
     """
     s = value_str.strip()
     lower = s.lower()
@@ -117,9 +119,11 @@ def _convert_value(value_str: str) -> Any:
     if not _LEADING_ZERO_RE.match(s):
         for converter in (int, float):
             try:
-                return converter(s)
+                number = converter(s)
             except ValueError:
                 continue
+            # float() also reads "nan" and "inf", which no JSON request can carry
+            return number if isinstance(number, int) or math.isfinite(number) else s
 
     # JSON handles lists, dicts, and explicitly-quoted strings
     try:
diff --git a/reme/steps/index/bm25_search.py b/reme/steps/index/bm25_search.py
index 3ac139c3..99ca8a4d 100644
--- a/reme/steps/index/bm25_search.py
+++ b/reme/steps/index/bm25_search.py
@@ -23,7 +23,8 @@ class Bm25SearchStep(_ToolContextDedupMixin, BaseStep):
 
     async def execute(self):
         assert self.context is not None
-        query: str = (self.context.get("query", "") or "").strip()
+        raw_query = self.context.get("query")
+        query: str = "" if raw_query is None else str(raw_query).strip()
         limit: int = int(self.context.get("limit") or 5)
         min_score: float = float(self.context.get("min_score") or 0.0)
         tool_context_id: str = (self.context.get("tool_context_id", "") or "").strip()
diff --git a/reme/steps/index/node_search.py b/reme/steps/index/node_search.py
index ae35e609..d6b50658 100644
--- a/reme/steps/index/node_search.py
+++ b/reme/steps/index/node_search.py
@@ -61,7 +61,8 @@ class NodeSearchStep(BaseStep):
 
     async def execute(self):
         assert self.context is not None
-        query: str = (self.context.get("query", "") or "").strip()
+        raw_query = self.context.get("query")
+        query: str = "" if raw_query is None else str(raw_query).strip()
         limit: int = int(self.context.get("limit") or 20)
         vector_weight: float = float(self.kwargs.get("vector_weight", 0.7))
         candidate_multiplier: float = float(self.kwargs.get("candidate_multiplier", 5.0))
diff --git a/reme/steps/index/search.py b/reme/steps/index/search.py
index b5aafd3b..8d644aa1 100644
--- a/reme/steps/index/search.py
+++ b/reme/steps/index/search.py
@@ -203,7 +203,8 @@ class SearchStep(BaseStep):
 
     async def execute(self):
         assert self.context is not None
-        query: str = (self.context.get("query", "") or "").strip()
+        raw_query = self.context.get("query")
+        query: str = "" if raw_query is None else str(raw_query).strip()
         limit: int = int(self.context.get("limit") or _default_limit())
         min_score: float = float(self.context.get("min_score") or 0.0)
         # vector_weight: prefer agent-supplied context value; fallback to YAML kwargs / default 0.7.
diff --git a/reme/steps/index/vector_search.py b/reme/steps/index/vector_search.py
index 638344e1..5de5aa94 100644
--- a/reme/steps/index/vector_search.py
+++ b/reme/steps/index/vector_search.py
@@ -23,7 +23,8 @@ class VectorSearchStep(_ToolContextDedupMixin, BaseStep):
 
     async def execute(self):
         assert self.context is not None
-        query: str = (self.context.get("query", "") or "").strip()
+        raw_query = self.context.get("query")
+        query: str = "" if raw_query is None else str(raw_query).strip()
         limit: int = int(self.context.get("limit") or 5)
         min_score: float = float(self.context.get("min_score") or 0.0)
         tool_context_id: str = (self.context.get("tool_context_id", "") or "").strip()
diff --git a/tests/unit/test_config_parser.py b/tests/unit/test_config_parser.py
index 1176014e..35b1c174 100644
--- a/tests/unit/test_config_parser.py
+++ b/tests/unit/test_config_parser.py
@@ -165,6 +165,16 @@ def test_parse_args_separates_action_and_application_kwargs():
     assert kwargs == {"query": "hello", "limit": 3}
 
 
+@pytest.mark.parametrize("value", ["nan", "NaN", "inf", "-Infinity", "1e999"])
+def test_parse_kwargs_keeps_non_finite_numbers_as_strings(value):
+    """JSON has no NaN or Infinity, so these values stay text instead of becoming floats."""
+    assert parse_kwargs(f"query={value}", "limit=5", "min_score=0.5") == {
+        "query": value,
+        "limit": 5,
+        "min_score": 0.5,
+    }
+
+
 @pytest.mark.parametrize("item", ["=1", ".a=1", "a.=1", "a..b=1"])
 def test_parse_dot_notation_rejects_empty_key_segments(item):
     """Dot notation keys cannot contain empty path segments."""
diff --git a/tests/unit/test_search_step.py b/tests/unit/test_search_step.py
index 5e218e4d..886ff67d 100644
--- a/tests/unit/test_search_step.py
+++ b/tests/unit/test_search_step.py
@@ -6,6 +6,7 @@ import asyncio
 import importlib.util
 from pathlib import Path
 
+import pytest
 from agentscope.message import Msg
 
 from reme.components.file_store import BaseFileStore
@@ -13,23 +14,32 @@ from reme.components.file_store.local_file_store import LocalFileStore
 from reme.components.tag_index import LocalTagIndex
 from reme.components import ApplicationContext
 from reme.components.runtime_context import RuntimeContext
+from reme.config.config_parser import parse_kwargs
 from reme.enumeration import LinkScopeEnum
 from reme.schema import FileChunk, FileLink, FileNode, Response
 from reme.steps.index import (
     AddDraftStep,
     Bm25SearchStep,
+    NodeSearchStep,
     ReadAllDraftStep,
     SearchStep,
     VectorSearchStep,
 )
 from reme.steps.index._source_format import ALL_RETURNED_MESSAGE, NO_RESULTS_MESSAGE
 
-_SEARCH_V2_PATH = Path(__file__).parents[2] / "plugins" / "beam" / "src" / "reme_beam" / "search_v2.py"
-_SEARCH_V2_SPEC = importlib.util.spec_from_file_location("reme_beam.search_v2", _SEARCH_V2_PATH)
-assert _SEARCH_V2_SPEC is not None and _SEARCH_V2_SPEC.loader is not None
-_SEARCH_V2_MODULE = importlib.util.module_from_spec(_SEARCH_V2_SPEC)
-_SEARCH_V2_SPEC.loader.exec_module(_SEARCH_V2_MODULE)
-SearchV2Step = _SEARCH_V2_MODULE.SearchV2Step
+
+def _load_search_v2_step(plugin: str) -> type:
+    """Load SearchV2Step from a benchmark plugin's source file."""
+    path = Path(__file__).parents[2] / "plugins" / plugin / "src" / f"reme_{plugin}" / "search_v2.py"
+    spec = importlib.util.spec_from_file_location(f"reme_{plugin}.search_v2", path)
+    assert spec is not None and spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module.SearchV2Step
+
+
+SearchV2Step = _load_search_v2_step("beam")
+LmeSearchV2Step = _load_search_v2_step("lme")
 
 
 class FakeSearchStore(BaseFileStore):
@@ -803,6 +813,36 @@ def test_search_steps_no_results_shows_no_results_message():
     asyncio.run(run())
 
 
+_QUERY_STEP_CLASSES = [
+    pytest.param(SearchStep, id="search"),
+    pytest.param(VectorSearchStep, id="vector_search"),
+    pytest.param(Bm25SearchStep, id="bm25_search"),
+    pytest.param(NodeSearchStep, id="node_search"),
+    pytest.param(SearchV2Step, id="beam_search_v2"),
+    pytest.param(LmeSearchV2Step, id="lme_search_v2"),
+]
+
+
+@pytest.mark.parametrize("value", ["404", "0", "0.0"])
+@pytest.mark.parametrize("step_cls", _QUERY_STEP_CLASSES)
+def test_search_steps_search_for_the_text_of_a_numeric_query(step_cls, value):
+    """The CLI parses ``query=0`` as a number; the search steps still search for its text."""
+
+    async def run():
+        store = FakeSearchStore(
+            vector_results=[_chunk("hit", "digest/a.md", f"Error {value} from the gateway", "vector", 0.9)],
+            keyword_results=[_chunk("hit", "digest/a.md", f"Error {value} from the gateway", "keyword", 4.0)],
+        )
+        query = parse_kwargs(f"query={value}")["query"]
+        resp = await step_cls(file_store=store)(RuntimeContext(query=query, limit=5))
+
+        assert resp.success is True
+        assert store.calls and {call[1] for call in store.calls} == {value}
+        assert "digest/a.md" in resp.answer
+
+    asyncio.run(run())
+
+
 # ---------------------------------------------------------------------------
 # SearchStep tests — exercise the SearchStep (simple chunk.id dedup,
 # inline answer formatting).