ouroboros-ai 0.40.2.dev12__tar.gz → 0.40.2.dev13__tar.gz

{ouroboros_ai-0.40.2.dev12 → ouroboros_ai-0.40.2.dev13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ouroboros-ai
-Version: 0.40.2.dev12
+Version: 0.40.2.dev13
 Summary: Specification-first workflow engine for AI coding agents. Works with Claude Code and Codex CLI.
 Author-email: Q00 <jqyu.lee@gmail.com>
 License-File: LICENSE

{ouroboros_ai-0.40.2.dev12 → ouroboros_ai-0.40.2.dev13}/src/ouroboros/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.40.2.dev12'
-__version_tuple__ = version_tuple = (0, 40, 2, 'dev12')
+__version__ = version = '0.40.2.dev13'
+__version_tuple__ = version_tuple = (0, 40, 2, 'dev13')
 
 __commit_id__ = commit_id = None

{ouroboros_ai-0.40.2.dev12 → ouroboros_ai-0.40.2.dev13}/src/ouroboros/cli/commands/setup.py

@@ -49,6 +49,12 @@ from ouroboros.cli.opencode_config import (
 from ouroboros.cli.opencode_config import (
     is_bridge_plugin_entry as _is_bridge_plugin_entry,
 )
+from ouroboros.config._model_defaults import (
+    DEFAULT_CONSENSUS_OPUS_MODEL,
+    DEFAULT_OPUS_MODEL,
+    DEFAULT_SONNET_MODEL,
+    recognized_shipped_defaults,
+)
 from ouroboros.persistence.brownfield import BrownfieldStore
 
 
@@ -324,38 +330,36 @@ _CODEX_DEFAULT_LLM_ROLE_PROFILES: dict[str, str] = {
 
 _DEFAULT_CONSENSUS_MODELS = (
     "openrouter/openai/gpt-4o",
-    "openrouter/anthropic/claude-opus-4-6",
+    DEFAULT_CONSENSUS_OPUS_MODEL,
     "openrouter/google/gemini-2.5-pro",
 )
 
 _MISSING = object()
 
 _CODEX_ROLE_MODEL_OVERRIDE_DEFAULTS: dict[str, tuple[tuple[tuple[str, ...], object], ...]] = {
-    "ambiguity": ((("clarification", "default_model"), "claude-opus-4-6"),),
-    "assertion_extraction": ((("evaluation", "assertion_extraction_model"), "claude-sonnet-4-6"),),
-    "atomicity": ((("execution", "atomicity_model"), "claude-opus-4-6"),),
-    "brownfield_explore": ((("clarification", "default_model"), "claude-opus-4-6"),),
-    "clarification": ((("clarification", "default_model"), "claude-opus-4-6"),),
-    "consensus_advocate": (
-        (("consensus", "advocate_model"), "openrouter/anthropic/claude-opus-4-6"),
-    ),
+    "ambiguity": ((("clarification", "default_model"), DEFAULT_OPUS_MODEL),),
+    "assertion_extraction": ((("evaluation", "assertion_extraction_model"), DEFAULT_SONNET_MODEL),),
+    "atomicity": ((("execution", "atomicity_model"), DEFAULT_OPUS_MODEL),),
+    "brownfield_explore": ((("clarification", "default_model"), DEFAULT_OPUS_MODEL),),
+    "clarification": ((("clarification", "default_model"), DEFAULT_OPUS_MODEL),),
+    "consensus_advocate": ((("consensus", "advocate_model"), DEFAULT_CONSENSUS_OPUS_MODEL),),
     "consensus_judge": ((("consensus", "judge_model"), "openrouter/google/gemini-2.5-pro"),),
     "consensus_vote": ((("consensus", "models"), _DEFAULT_CONSENSUS_MODELS),),
     "context_compression": ((("llm", "context_compression_model"), "gpt-4"),),
-    "decomposition": ((("execution", "decomposition_model"), "claude-opus-4-6"),),
-    "dependency_analysis": ((("llm", "dependency_analysis_model"), "claude-opus-4-6"),),
-    "double_diamond": ((("execution", "double_diamond_model"), "claude-opus-4-6"),),
-    "mechanical_detection": ((("evaluation", "assertion_extraction_model"), "claude-sonnet-4-6"),),
+    "decomposition": ((("execution", "decomposition_model"), DEFAULT_OPUS_MODEL),),
+    "dependency_analysis": ((("llm", "dependency_analysis_model"), DEFAULT_OPUS_MODEL),),
+    "double_diamond": ((("execution", "double_diamond_model"), DEFAULT_OPUS_MODEL),),
+    "mechanical_detection": ((("evaluation", "assertion_extraction_model"), DEFAULT_SONNET_MODEL),),
     "ontology_analysis": (
-        (("llm", "ontology_analysis_model"), "claude-opus-4-6"),
+        (("llm", "ontology_analysis_model"), DEFAULT_OPUS_MODEL),
         (("consensus", "devil_model"), "openrouter/openai/gpt-4o"),
     ),
-    "pm_interview": ((("clarification", "default_model"), "claude-opus-4-6"),),
-    "qa": ((("llm", "qa_model"), "claude-sonnet-4-6"),),
-    "reflect": ((("resilience", "reflect_model"), "claude-opus-4-6"),),
-    "seed_generation": ((("clarification", "default_model"), "claude-opus-4-6"),),
-    "semantic_evaluation": ((("evaluation", "semantic_model"), "claude-opus-4-6"),),
-    "wonder": ((("resilience", "wonder_model"), "claude-opus-4-6"),),
+    "pm_interview": ((("clarification", "default_model"), DEFAULT_OPUS_MODEL),),
+    "qa": ((("llm", "qa_model"), DEFAULT_SONNET_MODEL),),
+    "reflect": ((("resilience", "reflect_model"), DEFAULT_OPUS_MODEL),),
+    "seed_generation": ((("clarification", "default_model"), DEFAULT_OPUS_MODEL),),
+    "semantic_evaluation": ((("evaluation", "semantic_model"), DEFAULT_OPUS_MODEL),),
+    "wonder": ((("resilience", "wonder_model"), DEFAULT_OPUS_MODEL),),
 }
 
 
@@ -1588,19 +1592,19 @@ def _register_copilot_mcp_server() -> None:
 
 
 _COPILOT_DEFAULT_MODEL_TARGETS: tuple[tuple[str, str, str], ...] = (
-    ("llm", "qa_model", "claude-sonnet-4-6"),
-    ("llm", "dependency_analysis_model", "claude-opus-4-6"),
-    ("llm", "ontology_analysis_model", "claude-opus-4-6"),
+    ("llm", "qa_model", DEFAULT_SONNET_MODEL),
+    ("llm", "dependency_analysis_model", DEFAULT_OPUS_MODEL),
+    ("llm", "ontology_analysis_model", DEFAULT_OPUS_MODEL),
     ("llm", "context_compression_model", "gpt-4"),
-    ("clarification", "default_model", "claude-opus-4-6"),
-    ("evaluation", "semantic_model", "claude-opus-4-6"),
-    ("evaluation", "assertion_extraction_model", "claude-sonnet-4-6"),
-    ("resilience", "wonder_model", "claude-opus-4-6"),
-    ("resilience", "reflect_model", "claude-opus-4-6"),
-    ("execution", "atomicity_model", "claude-opus-4-6"),
-    ("execution", "decomposition_model", "claude-opus-4-6"),
-    ("execution", "double_diamond_model", "claude-opus-4-6"),
-    ("consensus", "advocate_model", "openrouter/anthropic/claude-opus-4-6"),
+    ("clarification", "default_model", DEFAULT_OPUS_MODEL),
+    ("evaluation", "semantic_model", DEFAULT_OPUS_MODEL),
+    ("evaluation", "assertion_extraction_model", DEFAULT_SONNET_MODEL),
+    ("resilience", "wonder_model", DEFAULT_OPUS_MODEL),
+    ("resilience", "reflect_model", DEFAULT_OPUS_MODEL),
+    ("execution", "atomicity_model", DEFAULT_OPUS_MODEL),
+    ("execution", "decomposition_model", DEFAULT_OPUS_MODEL),
+    ("execution", "double_diamond_model", DEFAULT_OPUS_MODEL),
+    ("consensus", "advocate_model", DEFAULT_CONSENSUS_OPUS_MODEL),
     ("consensus", "devil_model", "openrouter/openai/gpt-4o"),
     ("consensus", "judge_model", "openrouter/google/gemini-2.5-pro"),
 )
@@ -1611,7 +1615,7 @@ _COPILOT_DEFAULT_MODEL_LIST_TARGETS: tuple[tuple[str, str, tuple[str, ...]], ...
         "models",
         (
             "openrouter/openai/gpt-4o",
-            "openrouter/anthropic/claude-opus-4-6",
+            DEFAULT_CONSENSUS_OPUS_MODEL,
             "openrouter/google/gemini-2.5-pro",
         ),
     ),
@@ -1629,22 +1633,48 @@ def _apply_copilot_default_model(
     Treat setup's selected model as the default for model fields that are
     absent or still equal to Ouroboros' shipped defaults, while preserving
     explicit user overrides.
+
+    "Shipped default" means the *current* pin OR any prior-release shipped
+    default (``recognized_shipped_defaults``): a config persisted before a pin
+    bump holds the old literal but is still an untouched shipped default the
+    user never chose, so Copilot setup must rewrite it to the discovered model
+    rather than mistaking it for an explicit override (Q00/ouroboros#1324).
     """
     for section_name, key, shipped_default in _COPILOT_DEFAULT_MODEL_TARGETS:
         section = _ensure_mapping_section(config_dict, section_name)
         current = section.get(key)
-        if current is None or current == shipped_default:
+        if current is None or current in recognized_shipped_defaults(shipped_default):
             section[key] = chosen_model
 
     for section_name, key, shipped_default in _COPILOT_DEFAULT_MODEL_LIST_TARGETS:
         section = _ensure_mapping_section(config_dict, section_name)
         current = section.get(key)
         if current is None or (
-            isinstance(current, (list, tuple)) and tuple(current) == shipped_default
+            isinstance(current, (list, tuple))
+            and _is_shipped_default_roster(current, shipped_default)
         ):
             section[key] = list(model_roster)
 
 
+def _is_shipped_default_roster(
+    current: list | tuple,
+    shipped_roster: tuple[str, ...],
+) -> bool:
+    """True when ``current`` is a shipped default roster (current or legacy).
+
+    Element-wise match against ``recognized_shipped_defaults`` so a roster
+    persisted before a pin bump (e.g. the old OpenRouter Opus slug in the
+    consensus slot) is still recognized as an untouched shipped default.
+    """
+    current_tuple = tuple(current)
+    if len(current_tuple) != len(shipped_roster):
+        return False
+    return all(
+        str(candidate) in recognized_shipped_defaults(default)
+        for candidate, default in zip(current_tuple, shipped_roster, strict=True)
+    )
+
+
 def _setup_copilot(copilot_path: str, *, non_interactive: bool = False) -> None:
     """Configure Ouroboros for the GitHub Copilot CLI runtime.
 

ouroboros_ai-0.40.2.dev13/src/ouroboros/config/_model_defaults.py

@@ -0,0 +1,70 @@
+"""Single source of truth for default Claude model pins.
+
+Every config default (Pydantic field defaults in :mod:`ouroboros.config.models`,
+the ``get_*_model`` fallbacks in :mod:`ouroboros.config.loader`, and the setup
+wizard tables in :mod:`ouroboros.cli.commands.setup`) references the constants
+below. Bumping a pinned model is therefore a one-line edit here instead of
+shotgun surgery across three layers. See Q00/ouroboros#1322.
+
+From the Claude 4.6 generation onward, model IDs are dateless but still pinned
+snapshots (not evergreen pointers), so pinning remains fully reproducible:
+https://platform.claude.com/docs/en/about-claude/models/overview
+
+These pins intentionally do NOT use the ``"default"`` sentinel: the evaluation
+and consensus phases depend on a stable model tier for reproducible grading.
+
+Scope: these constants cover the Anthropic-direct API ids (used by the
+``claude``/``litellm`` backends) and the OpenRouter consensus roster. The
+Copilot setup path selects models from GitHub Copilot's own discovery catalog
+(distinct dotted ids surfaced by ``copilot.model_discovery``) and is therefore
+out of scope here — it is not driven by these pins.
+
+Note on id formats across providers (they are NOT interchangeable):
+- Anthropic direct API uses hyphenated, dateless ids: ``claude-opus-4-8``.
+- OpenRouter uses dotted slugs: ``anthropic/claude-opus-4.8``
+  (https://openrouter.ai/anthropic/claude-opus-4.8).
+"""
+
+# Frontier reasoning tier (interview, seed, ontology, evaluation, execution
+# analysis, consensus advocate). Anthropic-direct API id. Bump on each new
+# Opus release.
+DEFAULT_OPUS_MODEL = "claude-opus-4-8"
+
+# Speed/judgment tier (QA verdicts, assertion extraction). Bump on each new
+# Sonnet release.
+DEFAULT_SONNET_MODEL = "claude-sonnet-4-6"
+
+# OpenRouter-routed Opus for the multi-provider consensus roster. This is the
+# OpenRouter slug (dotted ``claude-opus-4.8``), which differs from the
+# Anthropic-direct id above — LiteLLM forwards it verbatim to OpenRouter, so it
+# must match OpenRouter's published model id exactly or consensus voting fails.
+DEFAULT_CONSENSUS_OPUS_MODEL = "openrouter/anthropic/claude-opus-4.8"
+
+
+# Historical shipped default pins from prior releases, keyed by the *current*
+# default that replaced them. A config persisted before a pin was bumped still
+# contains the older literal (e.g. ``claude-opus-4-6`` was the frozen Opus
+# default from 2026-02-28 until this change; ``claude-sonnet-4-20250514`` was
+# the pre-EOL QA default). Backends that cannot run Claude model names
+# (Codex/Copilot/Hermes/Kiro) must treat these legacy shipped defaults exactly
+# like the current shipped default and normalize them to the ``"default"``
+# sentinel — otherwise bumping a pin silently reclassifies an untouched shipped
+# default in an already-persisted config as an explicit user override and leaks
+# a Claude id to a backend that cannot execute it (Q00/ouroboros#1324 review).
+LEGACY_DEFAULT_MODELS: dict[str, tuple[str, ...]] = {
+    DEFAULT_OPUS_MODEL: ("claude-opus-4-6",),
+    DEFAULT_SONNET_MODEL: ("claude-sonnet-4-20250514",),
+    DEFAULT_CONSENSUS_OPUS_MODEL: ("openrouter/anthropic/claude-opus-4-6",),
+}
+
+
+def recognized_shipped_defaults(default_model: str) -> tuple[str, ...]:
+    """Return every shipped-default value (current + historical) for a pin.
+
+    The loader's backend normalization treats any of these as "the shipped
+    default the user never chose," so a persisted config from a prior release
+    still resolves to the backend-safe ``"default"`` sentinel after a pin bump.
+    Genuinely explicit, never-shipped model ids are absent from this set and so
+    remain preserved verbatim.
+    """
+    return (default_model, *LEGACY_DEFAULT_MODELS.get(default_model, ()))

{ouroboros_ai-0.40.2.dev12 → ouroboros_ai-0.40.2.dev13}/src/ouroboros/config/loader.py

@@ -48,6 +48,12 @@ from pydantic import ValidationError as PydanticValidationError
 import yaml
 
 from ouroboros.backends import get_backend_capability
+from ouroboros.config._model_defaults import (  # noqa: E402
+    DEFAULT_CONSENSUS_OPUS_MODEL,
+    DEFAULT_OPUS_MODEL,
+    DEFAULT_SONNET_MODEL,
+    recognized_shipped_defaults,
+)
 from ouroboros.config.models import (  # noqa: E402
     CredentialsConfig,
     OuroborosConfig,
@@ -72,10 +78,10 @@ _PLACEHOLDER_API_KEY_SUFFIX = "_API_KEY"
 _DEFAULT_MAX_PARALLEL_WORKERS = 3
 _DEFAULT_CONSENSUS_MODELS = (
     "openrouter/openai/gpt-4o",
-    "openrouter/anthropic/claude-opus-4-6",
+    DEFAULT_CONSENSUS_OPUS_MODEL,
     "openrouter/google/gemini-2.5-pro",
 )
-_DEFAULT_CONSENSUS_ADVOCATE_MODEL = "openrouter/anthropic/claude-opus-4-6"
+_DEFAULT_CONSENSUS_ADVOCATE_MODEL = DEFAULT_CONSENSUS_OPUS_MODEL
 _DEFAULT_CONSENSUS_DEVIL_MODEL = "openrouter/openai/gpt-4o"
 _DEFAULT_CONSENSUS_JUDGE_MODEL = "openrouter/google/gemini-2.5-pro"
 _DEFAULT_USAGE_LIMIT_PAUSE_HOURS = 5.0
@@ -1271,13 +1277,19 @@ def _normalize_configured_model_for_backend(
         return _default_model_for_backend(default_model, backend=backend)
 
     resolved = _resolve_llm_backend_for_models(backend)
-    if resolved in _CODEX_LLM_BACKENDS and candidate == default_model:
+    # Recognize the current shipped default AND prior-release shipped defaults
+    # (#1324): a config persisted before a pin bump still holds the old literal,
+    # and for Claude-incapable backends it must normalize to the sentinel just
+    # like the current default would. Genuinely explicit, never-shipped ids are
+    # absent from this set and fall through to be preserved verbatim.
+    is_shipped_default = candidate in recognized_shipped_defaults(default_model)
+    if resolved in _CODEX_LLM_BACKENDS and is_shipped_default:
         return _CODEX_DEFAULT_MODEL
-    if resolved in _KIRO_LLM_BACKENDS and candidate == default_model:
+    if resolved in _KIRO_LLM_BACKENDS and is_shipped_default:
         return _KIRO_DEFAULT_MODEL
-    if resolved in _COPILOT_LLM_BACKENDS and candidate == default_model:
+    if resolved in _COPILOT_LLM_BACKENDS and is_shipped_default:
         return _COPILOT_DEFAULT_MODEL
-    if resolved in _HERMES_LLM_BACKENDS and candidate == default_model:
+    if resolved in _HERMES_LLM_BACKENDS and is_shipped_default:
         return _HERMES_DEFAULT_MODEL
 
     return candidate
@@ -1294,10 +1306,19 @@ def _normalize_configured_models_for_backend(
     if not normalized:
         return _default_models_for_backend(default_models, backend=backend)
 
+    # Match the shipped roster element-wise against current + legacy shipped
+    # defaults (#1324), so a roster persisted before a pin bump (e.g. the old
+    # OpenRouter Opus slug in the consensus slot) still normalizes to the
+    # backend-safe sentinel for Claude-incapable backends instead of leaking an
+    # unrunnable id.
+    is_shipped_roster = len(normalized) == len(default_models) and all(
+        candidate in recognized_shipped_defaults(default)
+        for candidate, default in zip(normalized, default_models, strict=True)
+    )
     if (
         _resolve_llm_backend_for_models(backend)
         in (_CODEX_LLM_BACKENDS | _COPILOT_LLM_BACKENDS | _HERMES_LLM_BACKENDS)
-        and normalized == default_models
+        and is_shipped_roster
     ):
         return _default_models_for_backend(default_models, backend=backend)
 
@@ -1319,11 +1340,11 @@ def get_clarification_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.clarification.default_model,
-            default_model="claude-opus-4-6",
+            default_model=DEFAULT_OPUS_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-opus-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_OPUS_MODEL, backend=backend)
 
 
 def get_qa_model(backend: str | None = None) -> str:
@@ -1336,11 +1357,11 @@ def get_qa_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.llm.qa_model,
-            default_model="claude-sonnet-4-6",
+            default_model=DEFAULT_SONNET_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-sonnet-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_SONNET_MODEL, backend=backend)
 
 
 def get_dependency_analysis_model(backend: str | None = None) -> str:
@@ -1353,11 +1374,11 @@ def get_dependency_analysis_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.llm.dependency_analysis_model,
-            default_model="claude-opus-4-6",
+            default_model=DEFAULT_OPUS_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-opus-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_OPUS_MODEL, backend=backend)
 
 
 def get_ontology_analysis_model(backend: str | None = None) -> str:
@@ -1370,11 +1391,11 @@ def get_ontology_analysis_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.llm.ontology_analysis_model,
-            default_model="claude-opus-4-6",
+            default_model=DEFAULT_OPUS_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-opus-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_OPUS_MODEL, backend=backend)
 
 
 def get_context_compression_model(backend: str | None = None) -> str:
@@ -1404,11 +1425,11 @@ def get_atomicity_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.execution.atomicity_model,
-            default_model="claude-opus-4-6",
+            default_model=DEFAULT_OPUS_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-opus-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_OPUS_MODEL, backend=backend)
 
 
 def get_decomposition_model(backend: str | None = None) -> str:
@@ -1421,11 +1442,11 @@ def get_decomposition_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.execution.decomposition_model,
-            default_model="claude-opus-4-6",
+            default_model=DEFAULT_OPUS_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-opus-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_OPUS_MODEL, backend=backend)
 
 
 def get_double_diamond_model(backend: str | None = None) -> str:
@@ -1438,11 +1459,11 @@ def get_double_diamond_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.execution.double_diamond_model,
-            default_model="claude-opus-4-6",
+            default_model=DEFAULT_OPUS_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-opus-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_OPUS_MODEL, backend=backend)
 
 
 def get_wonder_model(backend: str | None = None) -> str:
@@ -1455,11 +1476,11 @@ def get_wonder_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.resilience.wonder_model,
-            default_model="claude-opus-4-6",
+            default_model=DEFAULT_OPUS_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-opus-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_OPUS_MODEL, backend=backend)
 
 
 def get_reflect_model(backend: str | None = None) -> str:
@@ -1472,11 +1493,11 @@ def get_reflect_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.resilience.reflect_model,
-            default_model="claude-opus-4-6",
+            default_model=DEFAULT_OPUS_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-opus-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_OPUS_MODEL, backend=backend)
 
 
 def get_semantic_model(backend: str | None = None) -> str:
@@ -1489,11 +1510,11 @@ def get_semantic_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.evaluation.semantic_model,
-            default_model="claude-opus-4-6",
+            default_model=DEFAULT_OPUS_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-opus-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_OPUS_MODEL, backend=backend)
 
 
 def get_assertion_extraction_model(backend: str | None = None) -> str:
@@ -1506,11 +1527,11 @@ def get_assertion_extraction_model(backend: str | None = None) -> str:
         config = load_config()
         return _normalize_configured_model_for_backend(
             config.evaluation.assertion_extraction_model,
-            default_model="claude-sonnet-4-6",
+            default_model=DEFAULT_SONNET_MODEL,
             backend=backend,
         )
     except ConfigError:
-        return _default_model_for_backend("claude-sonnet-4-6", backend=backend)
+        return _default_model_for_backend(DEFAULT_SONNET_MODEL, backend=backend)
 
 
 def get_mechanical_detector_model(backend: str | None = None) -> str:

{ouroboros_ai-0.40.2.dev12 → ouroboros_ai-0.40.2.dev13}/src/ouroboros/config/models.py

@@ -26,6 +26,11 @@ from typing import Any, Literal
 
 from pydantic import BaseModel, Field, field_validator
 
+from ouroboros.config._model_defaults import (
+    DEFAULT_CONSENSUS_OPUS_MODEL,
+    DEFAULT_OPUS_MODEL,
+    DEFAULT_SONNET_MODEL,
+)
 from ouroboros.orchestrator_stage import VALID_STAGE_KEYS
 
 
@@ -131,9 +136,9 @@ class LLMConfig(BaseModel, frozen=True):
     ] = "claude_code"
     permission_mode: Literal["default", "acceptEdits", "bypassPermissions"] = "default"
     opencode_permission_mode: Literal["default", "acceptEdits", "bypassPermissions"] = "acceptEdits"
-    qa_model: str = "claude-sonnet-4-6"
-    dependency_analysis_model: str = "claude-opus-4-6"
-    ontology_analysis_model: str = "claude-opus-4-6"
+    qa_model: str = DEFAULT_SONNET_MODEL
+    dependency_analysis_model: str = DEFAULT_OPUS_MODEL
+    ontology_analysis_model: str = DEFAULT_OPUS_MODEL
     context_compression_model: str = "gpt-4"
 
 
@@ -177,7 +182,7 @@ class ClarificationConfig(BaseModel, frozen=True):
     ambiguity_threshold: float = Field(default=0.2, ge=0.0, le=1.0)
     max_interview_rounds: int = Field(default=10, ge=1)
     model_tier: Literal["frugal", "standard", "frontier"] = "standard"
-    default_model: str = "claude-opus-4-6"
+    default_model: str = DEFAULT_OPUS_MODEL
 
 
 class ExecutionConfig(BaseModel, frozen=True):
@@ -193,9 +198,9 @@ class ExecutionConfig(BaseModel, frozen=True):
 
     max_iterations_per_ac: int = Field(default=10, ge=1)
     retrospective_interval: int = Field(default=3, ge=1)
-    atomicity_model: str = "claude-opus-4-6"
-    decomposition_model: str = "claude-opus-4-6"
-    double_diamond_model: str = "claude-opus-4-6"
+    atomicity_model: str = DEFAULT_OPUS_MODEL
+    decomposition_model: str = DEFAULT_OPUS_MODEL
+    double_diamond_model: str = DEFAULT_OPUS_MODEL
 
 
 class ResilienceConfig(BaseModel, frozen=True):
@@ -214,8 +219,8 @@ class ResilienceConfig(BaseModel, frozen=True):
     lateral_thinking_enabled: bool = True
     lateral_model_tier: Literal["frugal", "standard", "frontier"] = "frontier"
     lateral_temperature: float = Field(default=0.8, ge=0.0, le=2.0)
-    wonder_model: str = "claude-opus-4-6"
-    reflect_model: str = "claude-opus-4-6"
+    wonder_model: str = DEFAULT_OPUS_MODEL
+    reflect_model: str = DEFAULT_OPUS_MODEL
 
 
 class EvaluationConfig(BaseModel, frozen=True):
@@ -236,8 +241,8 @@ class EvaluationConfig(BaseModel, frozen=True):
     stage3_enabled: bool = True
     satisfaction_threshold: float = Field(default=0.8, ge=0.0, le=1.0)
     uncertainty_threshold: float = Field(default=0.3, ge=0.0, le=1.0)
-    semantic_model: str = "claude-opus-4-6"
-    assertion_extraction_model: str = "claude-sonnet-4-6"
+    semantic_model: str = DEFAULT_OPUS_MODEL
+    assertion_extraction_model: str = DEFAULT_SONNET_MODEL
 
 
 class ConsensusConfig(BaseModel, frozen=True):
@@ -258,10 +263,10 @@ class ConsensusConfig(BaseModel, frozen=True):
     diversity_required: bool = True
     models: tuple[str, ...] = (
         "openrouter/openai/gpt-4o",
-        "openrouter/anthropic/claude-opus-4-6",
+        DEFAULT_CONSENSUS_OPUS_MODEL,
         "openrouter/google/gemini-2.5-pro",
     )
-    advocate_model: str = "openrouter/anthropic/claude-opus-4-6"
+    advocate_model: str = DEFAULT_CONSENSUS_OPUS_MODEL
     devil_model: str = "openrouter/openai/gpt-4o"
     judge_model: str = "openrouter/google/gemini-2.5-pro"
 
@@ -610,7 +615,7 @@ def get_default_config() -> OuroborosConfig:
                     intelligence_range=(14, 16),
                     models=[
                         ModelConfig(provider="openai", model="gpt-4o"),
-                        ModelConfig(provider="anthropic", model="claude-sonnet-4-6"),
+                        ModelConfig(provider="anthropic", model=DEFAULT_SONNET_MODEL),
                         ModelConfig(provider="google", model="gemini-2.5-pro"),
                     ],
                     use_cases=["logic_design", "stage2_evaluation", "refactoring"],
@@ -620,7 +625,7 @@ def get_default_config() -> OuroborosConfig:
                     intelligence_range=(18, 20),
                     models=[
                         ModelConfig(provider="openai", model="o3"),
-                        ModelConfig(provider="anthropic", model="claude-opus-4-6"),
+                        ModelConfig(provider="anthropic", model=DEFAULT_OPUS_MODEL),
                     ],
                     use_cases=["consensus", "lateral_thinking", "big_bang"],
                 ),