ouroboros-ai 0.2.0__tar.gz → 0.2.2__tar.gz

{ouroboros_ai-0.2.0 → ouroboros_ai-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ouroboros-ai
-Version: 0.2.0
+Version: 0.2.2
 Summary: Self-Improving AI Workflow System
 Author-email: Q00 <jqyu.lee@gmail.com>
 License-File: LICENSE

{ouroboros_ai-0.2.0 → ouroboros_ai-0.2.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ouroboros-ai"
-version = "0.2.0"
+version = "0.2.2"
 description = "Self-Improving AI Workflow System"
 readme = "README.md"
 authors = [

{ouroboros_ai-0.2.0 → ouroboros_ai-0.2.2}/src/ouroboros/bigbang/ambiguity.py

@@ -36,8 +36,8 @@ DEFAULT_MODEL = "openrouter/google/gemini-2.0-flash-001"
 # Temperature for reproducible scoring
 SCORING_TEMPERATURE = 0.1
 
-# Maximum token limit to prevent cost explosion
-MAX_TOKEN_LIMIT = 8192
+# Maximum token limit (None = no limit, rely on model's context window)
+MAX_TOKEN_LIMIT: int | None = None
 
 
 class ComponentScore(BaseModel):
@@ -110,15 +110,15 @@ class AmbiguityScorer:
     from interview conversation, producing reproducible scores.
 
     Uses adaptive token allocation: starts with `initial_max_tokens` and
-    doubles on truncation up to `MAX_TOKEN_LIMIT`. Retries up to `max_retries`
-    times on both provider errors and parse failures.
+    doubles on truncation up to `MAX_TOKEN_LIMIT`. Retries until success
+    by default (unlimited), or up to `max_retries` if specified.
 
     Attributes:
         llm_adapter: The LLM adapter for completions.
         model: Model identifier to use.
         temperature: Temperature for reproducibility (default 0.1).
         initial_max_tokens: Starting token limit (default 2048).
-        max_retries: Maximum retry attempts (default 3).
+        max_retries: Maximum retry attempts, or None for unlimited (default).
 
     Example:
         scorer = AmbiguityScorer(llm_adapter=LiteLLMAdapter())
@@ -138,7 +138,7 @@ class AmbiguityScorer:
     model: str = DEFAULT_MODEL
     temperature: float = SCORING_TEMPERATURE
     initial_max_tokens: int = 2048
-    max_retries: int = 3
+    max_retries: int | None = None  # None = unlimited retries
 
     async def score(
         self, state: InterviewState
@@ -180,8 +180,15 @@ class AmbiguityScorer:
         current_max_tokens = self.initial_max_tokens
         last_error: Exception | ProviderError | None = None
         last_response: str = ""
+        attempt = 0
+
+        while True:
+            # Check retry limit if set
+            if self.max_retries is not None and attempt >= self.max_retries:
+                break
+
+            attempt += 1
 
-        for attempt in range(self.max_retries):
             config = CompletionConfig(
                 model=self.model,
                 temperature=self.temperature,
@@ -190,15 +197,15 @@ class AmbiguityScorer:
 
             result = await self.llm_adapter.complete(messages, config)
 
-            # Fix #3: Retry on provider errors (rate limits, transient failures)
+            # Retry on provider errors (rate limits, transient failures)
             if result.is_err:
                 last_error = result.error
                 log.warning(
                     "ambiguity.scoring.provider_error_retrying",
                     interview_id=state.interview_id,
                     error=str(result.error),
-                    attempt=attempt + 1,
-                    max_retries=self.max_retries,
+                    attempt=attempt,
+                    max_retries=self.max_retries or "unlimited",
                 )
                 continue
 
@@ -221,7 +228,7 @@ class AmbiguityScorer:
                     constraint_clarity=breakdown.constraint_clarity.clarity_score,
                     success_criteria_clarity=breakdown.success_criteria_clarity.clarity_score,
                     tokens_used=current_max_tokens,
-                    attempt=attempt + 1,
+                    attempt=attempt,
                 )
 
                 return Result.ok(ambiguity_score)
@@ -230,17 +237,19 @@ class AmbiguityScorer:
                 last_error = e
                 last_response = result.value.content
 
-                # Fix #2: Only increase tokens if response was truncated
+                # Only increase tokens if response was truncated
                 is_truncated = result.value.finish_reason == "length"
 
                 if is_truncated:
-                    # Fix #1: Cap token growth with MAX_TOKEN_LIMIT
-                    next_tokens = min(current_max_tokens * 2, MAX_TOKEN_LIMIT)
+                    # Double tokens on truncation, capped at MAX_TOKEN_LIMIT if set
+                    next_tokens = current_max_tokens * 2
+                    if MAX_TOKEN_LIMIT is not None:
+                        next_tokens = min(next_tokens, MAX_TOKEN_LIMIT)
                     log.warning(
                         "ambiguity.scoring.truncated_retrying",
                         interview_id=state.interview_id,
                         error=str(e),
-                        attempt=attempt + 1,
+                        attempt=attempt,
                         current_tokens=current_max_tokens,
                         next_tokens=next_tokens,
                     )
@@ -251,11 +260,11 @@ class AmbiguityScorer:
                         "ambiguity.scoring.format_error_retrying",
                         interview_id=state.interview_id,
                         error=str(e),
-                        attempt=attempt + 1,
+                        attempt=attempt,
                         finish_reason=result.value.finish_reason,
                     )
 
-        # All retries exhausted
+        # All retries exhausted (only reached if max_retries is set)
         log.warning(
             "ambiguity.scoring.failed",
             interview_id=state.interview_id,

{ouroboros_ai-0.2.0 → ouroboros_ai-0.2.2}/tests/unit/bigbang/test_ambiguity.py

@@ -302,7 +302,7 @@ class TestAmbiguityScorerInit:
         assert scorer.model == "openrouter/google/gemini-2.0-flash-001"
         assert scorer.temperature == SCORING_TEMPERATURE
         assert scorer.initial_max_tokens == 2048
-        assert scorer.max_retries == 3
+        assert scorer.max_retries is None  # Unlimited by default
 
     def test_scorer_custom_values(self) -> None:
         """AmbiguityScorer accepts custom values."""
@@ -1078,13 +1078,13 @@ class TestMaxTokenLimit:
     """Test MAX_TOKEN_LIMIT constant."""
 
     def test_max_token_limit_value(self) -> None:
-        """MAX_TOKEN_LIMIT prevents unbounded token growth."""
-        assert MAX_TOKEN_LIMIT == 8192
+        """MAX_TOKEN_LIMIT is None (no limit, rely on model's context window)."""
+        assert MAX_TOKEN_LIMIT is None
 
-    async def test_token_growth_capped_at_max_limit(self) -> None:
-        """Token growth is capped at MAX_TOKEN_LIMIT even with repeated truncation."""
+    async def test_token_growth_unbounded_when_no_limit(self) -> None:
+        """Token growth doubles without cap when MAX_TOKEN_LIMIT is None."""
         mock_adapter = MagicMock()
-        # All calls fail with truncation - tokens should cap at MAX_TOKEN_LIMIT
+        # All calls fail with truncation - tokens should keep doubling
         mock_adapter.complete = AsyncMock(
             return_value=Result.ok(
                 create_mock_completion_response(
@@ -1094,7 +1094,7 @@ class TestMaxTokenLimit:
             )
         )
 
-        # Start with 4096, should try 4096 -> 8192 -> 8192 (capped)
+        # Start with 4096, should try 4096 -> 8192 -> 16384 (no cap)
         scorer = AmbiguityScorer(
             llm_adapter=mock_adapter, initial_max_tokens=4096, max_retries=3
         )
@@ -1104,8 +1104,8 @@ class TestMaxTokenLimit:
 
         assert result.is_err  # All retries fail
         assert mock_adapter.complete.call_count == 3
-        # Verify token progression: 4096 -> 8192 -> 8192 (capped at MAX_TOKEN_LIMIT)
+        # Verify token progression: 4096 -> 8192 -> 16384 (no cap)
         configs = [call[0][1] for call in mock_adapter.complete.call_args_list]
         assert configs[0].max_tokens == 4096
         assert configs[1].max_tokens == 8192  # Doubled
-        assert configs[2].max_tokens == 8192  # Capped, not 16384
+        assert configs[2].max_tokens == 16384  # Doubled again, no cap

{ouroboros_ai-0.2.0 → ouroboros_ai-0.2.2}/uv.lock

@@ -892,7 +892,7 @@ wheels = [
 
 [[package]]
 name = "ouroboros-ai"
-version = "0.2.0"
+version = "0.2.1"
 source = { editable = "." }
 dependencies = [
     { name = "aiosqlite" },