fix(tools): ignore unusable output budget limits (#6516)

* fix(tools): ignore non-finite output budget limits

* fix(tools): skip overflowing budget estimates

README.md

@@ -315,6 +315,8 @@ streams simulate chunks from a non-streaming response, while no-tool streams sta
 
    OpenRouter and similar OpenAI-compatible gateways should be configured with `langchain_openai:ChatOpenAI` plus `base_url`. If you prefer a provider-specific environment variable name, point `api_key` at that variable explicitly (for example `api_key: $OPENROUTER_API_KEY`).
 
+   The `write_file` tool's output-budget hint uses the active model's effective `max_tokens`. Missing or unusable limits, including YAML `.inf`, `-.inf`, `.nan`, and values too large for the character estimate, omit this hint without failing tool assembly; provider-specific model validation still applies.
+
    To route OpenAI models through `/v1/responses`, keep using `langchain_openai:ChatOpenAI` and set `use_responses_api: true` with `output_version: responses/v1`.
 
    Models whose provider contract differs from DeerFlow's generic thinking/effort assumptions can declare a per-model mapping-valued `reasoning:` block (thinking `unsupported`/`optional`/`required`, the accepted effort values with aliases and a default, the payload dialect, and the reasoning-history requirement). The setup wizard's Z.AI GLM-5.3-Flash profile uses it: thinking stays on for every foreground and background call, and the effort selector offers the model's own `low`/`high`/`max` levels. Ollama's existing boolean `reasoning: true` remains a native provider setting and is forwarded to ChatOllama. When migrating a profile to a custom effort `path`, remove any old `reasoning_effort` setting from the profile and thinking templates; configuration validation rejects the leftover key. The chat UI drops a remembered provider-specific effort when switching to a legacy model that does not advertise it. Profiles without the block keep their existing provider behavior. See `config.example.yaml` for the shape and the equivalent manual configuration.

backend/packages/harness/deerflow/tools/AGENTS.md

@@ -23,7 +23,10 @@ the same host reader serves; keep reading guidance separate from permission enfo
 
 Lead and bootstrap assembly pass the constructed `chat_model` to tool assembly.
 The cloned `write_file` budget hint uses that instance's effective `max_tokens`,
-including custom-agent and thinking-mode overrides; an absent cap omits the hint.
+including custom-agent and thinking-mode overrides; an absent or unusable cap
+(including non-finite values or caps that overflow the character estimate)
+omits the hint without aborting tool assembly. Guard estimate arithmetic as
+well as integer conversion; a huge finite cap can pass the latter.
 Only standalone tool discovery without a model falls back to the base profile.
 
 `get_available_tools(groups, include_mcp, model_name, subagent_enabled)` assembles:

backend/packages/harness/deerflow/tools/tools.py

@@ -76,7 +76,7 @@ def _extract_max_tokens(model_config: object | None) -> int | None:
 
     Handles ModelConfig (where max_tokens may be stored as an extra dynamic field),
     dicts, SimpleNamespace, or test stubs. Rejects booleans, mocks, non-numeric
-    values, negative numbers, zero, and None.
+    values, non-finite numbers, negative numbers, zero, and None.
     """
     if model_config is None:
         return None
@@ -86,7 +86,7 @@ def _extract_max_tokens(model_config: object | None) -> int | None:
     try:
         val = int(raw)
         return val if val > 0 else None
-    except (ValueError, TypeError):
+    except (ValueError, TypeError, OverflowError):
         return None
 
 
@@ -219,22 +219,27 @@ def get_available_tools(
     # models configured without max_tokens.
     max_tokens = _extract_max_tokens(chat_model if chat_model is not None else model_config)
     if max_tokens is not None:
-        safe_chars = int(max_tokens * 3 * 0.7)
+        try:
-        budget_note = (
+            safe_chars = int(max_tokens * 3 * 0.7)
-            f"\n\nPER-RESPONSE BUDGET: your output limit is {max_tokens} tokens "
+        except OverflowError:
-            f"(≈{safe_chars} chars). Single non-append writes above this will be truncated. "
+            # Even finite caps can be too large for this optional estimate.
-            "For larger documents, write the first section now, "
+            pass
-            "then use append=True for subsequent sections."
+        else:
-        )
+            budget_note = (
-        loaded_tools = [
+                f"\n\nPER-RESPONSE BUDGET: your output limit is {max_tokens} tokens "
-            _clone_tool_with_description(
+                f"(≈{safe_chars} chars). Single non-append writes above this will be truncated. "
-                tool,
+                "For larger documents, write the first section now, "
-                f"{getattr(tool, 'description', '') or ''}{budget_note}",
+                "then use append=True for subsequent sections."
             )
-            if tool.name == "write_file" and hasattr(tool, "description") and "PER-RESPONSE BUDGET:" not in (getattr(tool, "description", "") or "")
+            loaded_tools = [
-            else tool
+                _clone_tool_with_description(
-            for tool in loaded_tools
+                    tool,
-        ]
+                    f"{getattr(tool, 'description', '') or ''}{budget_note}",
+                )
+                if tool.name == "write_file" and hasattr(tool, "description") and "PER-RESPONSE BUDGET:" not in (getattr(tool, "description", "") or "")
+                else tool
+                for tool in loaded_tools
+            ]
 
     # Get cached MCP tools if enabled
     # NOTE: We use ExtensionsConfig.from_file() instead of config.extensions

backend/tests/test_write_file_tool_budget_annotation.py

@@ -10,6 +10,7 @@ from types import SimpleNamespace
 from unittest.mock import MagicMock
 
 import pytest
+import yaml
 from pydantic import BaseModel
 
 from deerflow.config.app_config import AppConfig, ModelConfig, SandboxConfig, ToolConfig
@@ -70,6 +71,14 @@ def test_extract_max_tokens_various_inputs():
     assert _extract_max_tokens(mock_with) == 8000
 
 
+@pytest.mark.parametrize("max_tokens", [float("inf"), float("-inf"), float("nan")], ids=["inf", "negative-inf", "nan"])
+def test_extract_max_tokens_rejects_non_finite_values(max_tokens):
+    """Non-finite caps are unusable for every supported config representation."""
+    assert _extract_max_tokens({"max_tokens": max_tokens}) is None
+    assert _extract_max_tokens(SimpleNamespace(max_tokens=max_tokens)) is None
+    assert _extract_max_tokens(ModelConfig(name="test", model="m", use="u", max_tokens=max_tokens)) is None
+
+
 def test_clone_tool_with_description_preserves_singleton():
     """Verify _clone_tool_with_description returns an isolated copy and keeps original unchanged."""
     original_desc = write_file_tool.description
@@ -107,6 +116,34 @@ def test_get_available_tools_with_model_config_lacking_max_tokens():
     assert "PER-RESPONSE BUDGET:" not in write_tool.description
 
 
+@pytest.mark.parametrize(
+    "yaml_max_tokens",
+    [
+        ".inf",
+        "-.inf",
+        ".nan",
+        pytest.param("1.0e+308", id="huge-finite-float"),
+        pytest.param("1" + "0" * 400, id="huge-integer"),
+        pytest.param('"' + "1" + "0" * 400 + '"', id="huge-integer-string"),
+    ],
+)
+@pytest.mark.parametrize("use_chat_model", [False, True], ids=["profile", "constructed-model"])
+def test_get_available_tools_omits_unusable_budget(yaml_max_tokens, use_chat_model):
+    """Non-finite or overflowing caps must not abort assembly or produce a hint."""
+    baseline_desc = write_file_tool.description
+    max_tokens = yaml.safe_load(f"max_tokens: {yaml_max_tokens}")["max_tokens"]
+    model = ModelConfig(name="budget-model", model="m", use="u", max_tokens=4096 if use_chat_model else max_tokens)
+    config = _build_minimal_app_config([model])
+    chat_model = SimpleNamespace(max_tokens=max_tokens) if use_chat_model else None
+
+    tools = get_available_tools(model_name="budget-model", app_config=config, include_mcp=False, chat_model=chat_model)
+    write_tool = next(t for t in tools if t.name == "write_file")
+
+    assert "PER-RESPONSE BUDGET:" not in write_tool.description
+    assert write_tool.description == baseline_desc
+    assert write_file_tool.description == baseline_desc
+
+
 def test_write_file_singleton_remains_unmutated_across_assemblies():
     """Verify write_file_tool process singleton is never mutated during tool assembly."""
     baseline_desc = write_file_tool.description