fix(blocks): raise AIConditionBlock errors instead of swallowing them (#12593)

## Why

Sentry alert
[AUTOGPT-SERVER-8C8](https://significant-gravitas.sentry.io/issues/7367978095/)
— `AIConditionBlock` failing in prod with:

```
Invalid 'max_output_tokens': integer below minimum value.
Expected a value >= 16, but got 10 instead.
```

Two problems:
1. `max_tokens=10` is below OpenAI's new minimum of 16
2. The `except Exception` handler was calling `logger.error()` which
triggered Sentry for what are known block errors, AND silently
defaulting to `result=False` — making the block appear to succeed with
an incorrect answer

## What

- Bump `max_tokens` from 10 to 16 (fixes the root cause)
- Remove the `try/except` entirely — the executor already handles
exceptions correctly (`ValueError` = known/no Sentry, everything else =
unknown/Sentry). The old handler was just swallowing errors and
producing wrong results.

## Test plan

- [x] Existing `AIConditionBlock` tests pass (block only expects
"true"/"false", 16 tokens is plenty)
- [x] No more silent `result=False` on errors
- [x] No more spurious Sentry alerts from `logger.error()`

Fixes AUTOGPT-SERVER-8C8

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

autogpt_platform/backend/backend/blocks/ai_condition.py

@@ -1,3 +1,4 @@
+import re
 from typing import Any
 
 from backend.blocks._base import (
@@ -19,6 +20,33 @@ from backend.blocks.llm import (
 )
 from backend.data.model import APIKeyCredentials, NodeExecutionStats, SchemaField
 
+# Minimum max_output_tokens accepted by OpenAI-compatible APIs.
+# A true/false answer fits comfortably within this budget.
+MIN_LLM_OUTPUT_TOKENS = 16
+
+
+def _parse_boolean_response(response_text: str) -> tuple[bool, str | None]:
+    """Parse an LLM response into a boolean result.
+
+    Returns a ``(result, error)`` tuple.  *error* is ``None`` when the
+    response is unambiguous; otherwise it contains a diagnostic message
+    and *result* defaults to ``False``.
+    """
+    text = response_text.strip().lower()
+    if text == "true":
+        return True, None
+    if text == "false":
+        return False, None
+
+    # Fuzzy match – use word boundaries to avoid false positives like "untrue".
+    tokens = set(re.findall(r"\b(true|false|yes|no|1|0)\b", text))
+    if tokens == {"true"} or tokens == {"yes"} or tokens == {"1"}:
+        return True, None
+    if tokens == {"false"} or tokens == {"no"} or tokens == {"0"}:
+        return False, None
+
+    return False, f"Unclear AI response: '{response_text}'"
+
 
 class AIConditionBlock(AIBlockBase):
     """
@@ -162,54 +190,26 @@ class AIConditionBlock(AIBlockBase):
         ]
 
         # Call the LLM
-        try:
-            response = await self.llm_call(
-                credentials=credentials,
-                llm_model=input_data.model,
-                prompt=prompt,
-                max_tokens=10,  # We only expect a true/false response
+        response = await self.llm_call(
+            credentials=credentials,
+            llm_model=input_data.model,
+            prompt=prompt,
+            max_tokens=MIN_LLM_OUTPUT_TOKENS,
+        )
+
+        # Extract the boolean result from the response
+        result, error = _parse_boolean_response(response.response)
+        if error:
+            yield "error", error
+
+        # Update internal stats
+        self.merge_stats(
+            NodeExecutionStats(
+                input_token_count=response.prompt_tokens,
+                output_token_count=response.completion_tokens,
             )
-
-            # Extract the boolean result from the response
-            response_text = response.response.strip().lower()
-            if response_text == "true":
-                result = True
-            elif response_text == "false":
-                result = False
-            else:
-                # If the response is not clear, try to interpret it using word boundaries
-                import re
-
-                # Use word boundaries to avoid false positives like 'untrue' or '10'
-                tokens = set(re.findall(r"\b(true|false|yes|no|1|0)\b", response_text))
-
-                if tokens == {"true"} or tokens == {"yes"} or tokens == {"1"}:
-                    result = True
-                elif tokens == {"false"} or tokens == {"no"} or tokens == {"0"}:
-                    result = False
-                else:
-                    # Unclear or conflicting response - default to False and yield error
-                    result = False
-                    yield "error", f"Unclear AI response: '{response.response}'"
-
-            # Update internal stats
-            self.merge_stats(
-                NodeExecutionStats(
-                    input_token_count=response.prompt_tokens,
-                    output_token_count=response.completion_tokens,
-                )
-            )
-            self.prompt = response.prompt
-
-        except Exception as e:
-            # In case of any error, default to False to be safe
-            result = False
-            # Log the error but don't fail the block execution
-            import logging
-
-            logger = logging.getLogger(__name__)
-            logger.error(f"AI condition evaluation failed: {str(e)}")
-            yield "error", f"AI evaluation failed: {str(e)}"
+        )
+        self.prompt = response.prompt
 
         # Yield results
         yield "result", result

autogpt_platform/backend/backend/blocks/ai_condition_test.py

@@ -0,0 +1,147 @@
+"""Tests for AIConditionBlock – regression coverage for max_tokens and error propagation."""
+
+from __future__ import annotations
+
+from typing import cast
+
+import pytest
+
+from backend.blocks.ai_condition import (
+    MIN_LLM_OUTPUT_TOKENS,
+    AIConditionBlock,
+    _parse_boolean_response,
+)
+from backend.blocks.llm import (
+    DEFAULT_LLM_MODEL,
+    TEST_CREDENTIALS,
+    TEST_CREDENTIALS_INPUT,
+    AICredentials,
+    LLMResponse,
+)
+
+_TEST_AI_CREDENTIALS = cast(AICredentials, TEST_CREDENTIALS_INPUT)
+
+
+# ---------------------------------------------------------------------------
+# Helper to collect all yields from the async generator
+# ---------------------------------------------------------------------------
+
+
+async def _collect_outputs(block: AIConditionBlock, input_data, credentials):
+    outputs: dict[str, object] = {}
+    async for name, value in block.run(input_data, credentials=credentials):
+        outputs[name] = value
+    return outputs
+
+
+def _make_input(**overrides) -> AIConditionBlock.Input:
+    defaults: dict = {
+        "input_value": "hello@example.com",
+        "condition": "the input is an email address",
+        "yes_value": "yes!",
+        "no_value": "no!",
+        "model": DEFAULT_LLM_MODEL,
+        "credentials": TEST_CREDENTIALS_INPUT,
+    }
+    defaults.update(overrides)
+    return AIConditionBlock.Input(**defaults)
+
+
+def _mock_llm_response(response_text: str) -> LLMResponse:
+    return LLMResponse(
+        raw_response="",
+        prompt=[],
+        response=response_text,
+        tool_calls=None,
+        prompt_tokens=10,
+        completion_tokens=5,
+        reasoning=None,
+    )
+
+
+# ---------------------------------------------------------------------------
+# _parse_boolean_response unit tests
+# ---------------------------------------------------------------------------
+
+
+class TestParseBooleanResponse:
+    def test_true_exact(self):
+        assert _parse_boolean_response("true") == (True, None)
+
+    def test_false_exact(self):
+        assert _parse_boolean_response("false") == (False, None)
+
+    def test_true_with_whitespace(self):
+        assert _parse_boolean_response("  True  ") == (True, None)
+
+    def test_yes_fuzzy(self):
+        assert _parse_boolean_response("Yes") == (True, None)
+
+    def test_no_fuzzy(self):
+        assert _parse_boolean_response("no") == (False, None)
+
+    def test_one_fuzzy(self):
+        assert _parse_boolean_response("1") == (True, None)
+
+    def test_zero_fuzzy(self):
+        assert _parse_boolean_response("0") == (False, None)
+
+    def test_unclear_response(self):
+        result, error = _parse_boolean_response("I'm not sure")
+        assert result is False
+        assert error is not None
+        assert "Unclear" in error
+
+    def test_conflicting_tokens(self):
+        result, error = _parse_boolean_response("true and false")
+        assert result is False
+        assert error is not None
+
+
+# ---------------------------------------------------------------------------
+# Regression: max_tokens is set to MIN_LLM_OUTPUT_TOKENS
+# ---------------------------------------------------------------------------
+
+
+class TestMaxTokensRegression:
+    @pytest.mark.asyncio
+    async def test_llm_call_receives_min_output_tokens(self):
+        """max_tokens must be MIN_LLM_OUTPUT_TOKENS (16) – the previous value
+        of 1 was too low and caused OpenAI to reject the request."""
+        block = AIConditionBlock()
+        captured_kwargs: dict = {}
+
+        async def spy_llm_call(**kwargs):
+            captured_kwargs.update(kwargs)
+            return _mock_llm_response("true")
+
+        block.llm_call = spy_llm_call  # type: ignore[assignment]
+
+        input_data = _make_input()
+        await _collect_outputs(block, input_data, credentials=TEST_CREDENTIALS)
+
+        assert captured_kwargs["max_tokens"] == MIN_LLM_OUTPUT_TOKENS
+        assert captured_kwargs["max_tokens"] == 16
+
+
+# ---------------------------------------------------------------------------
+# Regression: exceptions from llm_call must propagate
+# ---------------------------------------------------------------------------
+
+
+class TestExceptionPropagation:
+    @pytest.mark.asyncio
+    async def test_llm_call_exception_propagates(self):
+        """If llm_call raises, the exception must NOT be swallowed.
+        Previously the block caught all exceptions and silently returned
+        result=False."""
+        block = AIConditionBlock()
+
+        async def boom(**kwargs):
+            raise RuntimeError("LLM provider error")
+
+        block.llm_call = boom  # type: ignore[assignment]
+
+        input_data = _make_input()
+        with pytest.raises(RuntimeError, match="LLM provider error"):
+            await _collect_outputs(block, input_data, credentials=TEST_CREDENTIALS)