Fix lifecycle CI: replace defunct Anthropic model IDs and honor CODEFRAME_LIFECYCLE_MODEL

codeframe/adapters/llm/base.py

@@ -44,11 +44,11 @@ class Purpose(str, Enum):
 
 
 # Default model aliases (use valid Anthropic model identifiers)
-DEFAULT_PLANNING_MODEL = "claude-sonnet-4-20250514"
-DEFAULT_EXECUTION_MODEL = "claude-sonnet-4-20250514"
-DEFAULT_GENERATION_MODEL = "claude-3-5-haiku-20241022"
-DEFAULT_CORRECTION_MODEL = "claude-opus-4-20250514"  # Step up for fixing errors
-DEFAULT_SUPERVISION_MODEL = "claude-opus-4-20250514"  # Strong model for supervisor decisions
+DEFAULT_PLANNING_MODEL = "claude-sonnet-4-5"
+DEFAULT_EXECUTION_MODEL = "claude-sonnet-4-5"
+DEFAULT_GENERATION_MODEL = "claude-haiku-4-5"
+DEFAULT_CORRECTION_MODEL = "claude-sonnet-4-5"  # Use same tier; override via CODEFRAME_CORRECTION_MODEL for a stronger model
+DEFAULT_SUPERVISION_MODEL = "claude-sonnet-4-5"  # Use same tier; override via CODEFRAME_SUPERVISION_MODEL for a stronger model
 
 
 @dataclass

codeframe/core/adapters/streaming_chat.py

@@ -35,7 +35,7 @@
 # Default model (matches DEFAULT_PLANNING_MODEL in base.py)
 # ---------------------------------------------------------------------------
 
-_DEFAULT_MODEL = "claude-sonnet-4-20250514"
+_DEFAULT_MODEL = "claude-sonnet-4-5"
 
 # Maximum token budget for conversation history passed to the API.
 # claude-sonnet-4 context window is 200k; leave 20k for response headroom.
@@ -465,15 +465,16 @@ async def _stream_turn(
 def _estimate_cost(input_tokens: int, output_tokens: int, model: str) -> float:
     """Rough cost estimate in USD.
 
-    Uses approximate pricing for claude-sonnet-4. Returns 0.0 for unknown models
+    Uses approximate pricing for claude-sonnet-4-5. Returns 0.0 for unknown models
     rather than raising — cost tracking is best-effort.
     """
     # Per-million-token pricing (input, output) in USD.
-    # Last verified: 2026-03-31. Anthropic pricing changes without notice —
+    # Last verified: 2026-06-21. Anthropic pricing changes without notice —
     # treat these as best-effort estimates, not billing-accurate figures.
     _PRICING: dict[str, tuple[float, float]] = {
-        "claude-sonnet-4-20250514": (3.0, 15.0),
-        "claude-opus-4-20250514": (15.0, 75.0),
+        "claude-sonnet-4-5": (3.0, 15.0),
+        "claude-opus-4-5": (15.0, 75.0),
+        "claude-haiku-4-5": (0.8, 4.0),
         "claude-3-5-haiku-20241022": (0.8, 4.0),
     }
     # Match by prefix to handle minor model variant suffixes

tests/lifecycle/conftest.py

@@ -18,6 +18,27 @@
 
 SAMPLE_PROJECT_DIR = Path(__file__).parent / "sample_project"
 
+# Map the CODEFRAME_LIFECYCLE_MODEL shorthand (haiku/sonnet) to actual model IDs.
+# These are passed to all `cf` subprocesses via per-purpose env vars so the agent
+# uses the requested model class instead of the compiled-in defaults.
+_LIFECYCLE_MODEL_MAP: dict[str, str] = {
+    "haiku": "claude-haiku-4-5",
+    "sonnet": "claude-sonnet-4-5",
+}
+
+
+def _lifecycle_model_env() -> dict[str, str]:
+    """Return per-purpose model env vars based on CODEFRAME_LIFECYCLE_MODEL."""
+    alias = os.getenv("CODEFRAME_LIFECYCLE_MODEL", "haiku")
+    model = _LIFECYCLE_MODEL_MAP.get(alias, "claude-haiku-4-5")
+    return {
+        "CODEFRAME_PLANNING_MODEL": model,
+        "CODEFRAME_EXECUTION_MODEL": model,
+        "CODEFRAME_GENERATION_MODEL": model,
+        "CODEFRAME_CORRECTION_MODEL": model,
+        "CODEFRAME_SUPERVISION_MODEL": model,
+    }
+
 
 def pytest_configure(config):
     config.addinivalue_line(
@@ -83,7 +104,7 @@ def run(*args, cwd=None, timeout=60, **kwargs):
             capture_output=True,
             text=True,
             timeout=timeout,
-            env={**os.environ},
+            env={**os.environ, **_lifecycle_model_env()},
         )
     return run