Switch optimizer and generator from Gemini to Claude

Migrates optimizer from Gemini 3.1 Pro to Claude Opus 4.6 and generator from Gemini Flash to Claude Sonnet 4.6. Updates client.py and slide_generator.py to use anthropic SDK instead of google-genai for their respective LLM calls. Adds anthropic>=0.40.0 to pyproject.toml dependencies. Updates log messages to reflect new model names.

openenv/client.py

@@ -26,18 +26,17 @@ import textwrap
 from pathlib import Path
 from typing import Any
 
+import anthropic
+import httpx
 from dotenv import load_dotenv
-from google import genai
+from loguru import logger
 
 load_dotenv(Path(__file__).parent.parent / ".env")
-from google.genai import types
-import httpx
-from loguru import logger
 
 from models import SlideSkillObservation
 
 SERVER_URL = "http://localhost:8000"
-OPTIMIZER_MODEL = "gemini-3.1-pro-preview"
+OPTIMIZER_MODEL = "claude-opus-4-6"
 
 BASELINE_EXAMPLES_CONTENT = "(Empty — no prior optimization rounds)\n"
 
@@ -64,7 +63,7 @@ class SlideSkillClient:
 
         Args:
             session_id: Active session ID.
-            action: Dict matching EditSectionAction or ReplaceFileAction schema.
+            action: Dict matching any SlideSkillAction schema.
                     Must include "action_type" key.
         """
         payload = {"session_id": session_id, "action": action}
@@ -94,15 +93,13 @@ class SlideSkillClient:
 
 def call_optimizer_llm(
     obs: SlideSkillObservation,
-    gemini_client: genai.Client,
+    anthropic_client: anthropic.Anthropic,
 ) -> dict[str, Any]:
     """
-    Call the optimizer LLM to generate a new DESIGN_RULES.md based on
+    Call Claude Opus 4.6 to generate a new DESIGN_RULES.md based on
     the evaluation feedback.
 
     Returns a dict suitable for the step() action parameter.
-    Uses ReplaceFileAction since the historical optimizer rewrites
-    the file wholesale.
     """
     prompt = textwrap.dedent(f"""\
         You are a McKinsey slide design optimizer. You are improving a
@@ -141,13 +138,13 @@ def call_optimizer_llm(
         no code fences.
     """)
 
-    response = gemini_client.models.generate_content(
+    response = anthropic_client.messages.create(
         model=OPTIMIZER_MODEL,
-        contents=prompt,
-        config=types.GenerateContentConfig(max_output_tokens=4096),
+        max_tokens=4096,
+        messages=[{"role": "user", "content": prompt}],
     )
 
-    new_content = response.text.strip()
+    new_content = response.content[0].text.strip()
 
     return {
         "action_type": "replace_file",
@@ -163,7 +160,7 @@ def run_optimization_loop(server_url: str = SERVER_URL, max_steps: int = 7) -> N
     This mirrors the historical Skill Forge loop but driven through the
     OpenEnv HTTP interface.
     """
-    gemini_client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])
+    anthropic_client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
 
     with SlideSkillClient(base_url=server_url) as client:
         logger.info(f"Starting optimization loop (max {max_steps} steps) | server={server_url}")
@@ -171,9 +168,9 @@ def run_optimization_loop(server_url: str = SERVER_URL, max_steps: int = 7) -> N
         logger.info(f"Session: {session_id}")
 
         # Step 0: baseline — generate slide with unmodified skill files.
-        logger.info("Step 0/baseline | generating slide (Flash)...")
+        logger.info("Step 0/baseline | generating slide (Sonnet 4.6)...")
         logger.info("Step 0/baseline | running Node.js + LibreOffice → JPG...")
-        logger.info("Step 0/baseline | evaluating slide (Pro)...")
+        logger.info("Step 0/baseline | evaluating slide (Gemini 3.1 Pro)...")
         obs = client.step(
             session_id,
             {
@@ -189,11 +186,11 @@ def run_optimization_loop(server_url: str = SERVER_URL, max_steps: int = 7) -> N
                 logger.info("Episode complete (max_steps reached).")
                 break
 
-            logger.info(f"Step {step_idx}/{max_steps} | optimizing skill files (Pro)...")
-            action = call_optimizer_llm(obs, gemini_client)
-            logger.info(f"Step {step_idx}/{max_steps} | generating slide (Flash)...")
+            logger.info(f"Step {step_idx}/{max_steps} | optimizing skill files (Opus 4.6)...")
+            action = call_optimizer_llm(obs, anthropic_client)
+            logger.info(f"Step {step_idx}/{max_steps} | generating slide (Sonnet 4.6)...")
             logger.info(f"Step {step_idx}/{max_steps} | running Node.js + LibreOffice → JPG...")
-            logger.info(f"Step {step_idx}/{max_steps} | evaluating slide (Pro)...")
+            logger.info(f"Step {step_idx}/{max_steps} | evaluating slide (Gemini 3.1 Pro)...")
             obs = client.step(session_id, action)
 
             delta_str = f"{obs.reward * 100:+.0f} pts"

openenv/slide_generator.py

@@ -2,7 +2,7 @@
 Slide Generator — orchestrates the full PPT generation pipeline.
 
 Pipeline (in order):
-    1. LLM reads skill files + TASK_PROMPT.md + js_templates + constraints
+    1. Claude Sonnet 4.6 reads skill files + TASK_PROMPT.md + js_templates + constraints
        → writes pptxgenjs JavaScript to generate.js in the session output dir.
     2. Code patches are applied to the generated JS.
     3. `node generate.js` runs in the session output dir → produces slide.pptx.
@@ -29,8 +29,7 @@ import subprocess
 import textwrap
 from pathlib import Path
 
-from google import genai
-from google.genai import types
+import anthropic
 
 from models import SlideSkillState
 
@@ -50,8 +49,8 @@ PDFTOPPM = (
     or "pdftoppm"
 )
 
-# Gemini Flash: fast and cost-effective for code generation.
-GENERATOR_MODEL = "gemini-3-flash-preview"
+# Claude Sonnet 4.6: fast and cost-effective for code generation.
+GENERATOR_MODEL = "claude-sonnet-4-6"
 GENERATOR_MAX_TOKENS = 4096
 
 
@@ -67,7 +66,7 @@ class SlideGenerator:
         self.task_prompt = task_prompt_path.read_text(encoding="utf-8")
         self.pptx_skill_dir = pptx_skill_dir
         self.reference_dir = reference_dir
-        self._client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])
+        self._client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
 
     def generate(
         self,
@@ -180,7 +179,7 @@ class SlideGenerator:
         node_error: str | None = None,
     ) -> str:
         """
-        Call the generator LLM with skill files + task prompt as context.
+        Call Claude Sonnet 4.6 with skill files + task prompt as context.
 
         Reads skill files from the session directory first (so agent edits
         take effect), falling back to the repo's pptx/ dir for any files
@@ -191,8 +190,6 @@ class SlideGenerator:
         design_rules = (session_dir / "DESIGN_RULES.md").read_text(encoding="utf-8")
         examples = (session_dir / "EXAMPLES.md").read_text(encoding="utf-8")
 
-        # Load pptx tooling from session dir (edited copies) with fallback
-        # to repo originals.
         pptx_skill = self._read_pptx_skill(session_dir)
 
         system_prompt = textwrap.dedent("""\
@@ -221,7 +218,6 @@ class SlideGenerator:
             {self.task_prompt}
         """)
 
-        # Inject hard constraints if any.
         if state and state.constraints:
             constraints_text = "\n".join(f"- {c}" for c in state.constraints)
             user_message += textwrap.dedent(f"""
@@ -230,7 +226,6 @@ class SlideGenerator:
             {constraints_text}
             """)
 
-        # Inject JS templates if any.
         if state and state.js_templates:
             templates_text = ""
             for name, code in state.js_templates.items():
@@ -254,27 +249,22 @@ class SlideGenerator:
             {node_error}
         """)
 
-        response = self._client.models.generate_content(
+        response = self._client.messages.create(
             model=GENERATOR_MODEL,
-            contents=user_message,
-            config=types.GenerateContentConfig(
-                system_instruction=system_prompt,
-                max_output_tokens=GENERATOR_MAX_TOKENS,
-            ),
+            max_tokens=GENERATOR_MAX_TOKENS,
+            system=system_prompt,
+            messages=[{"role": "user", "content": user_message}],
         )
 
-        code = response.text.strip()
+        code = response.content[0].text.strip()
 
-        # Extract from markdown code fence if present (LLMs often add them
-        # despite instructions). Handles ```javascript, ```js, or plain ```.
+        # Extract from markdown code fence if present.
         fence_match = re.search(r"```(?:javascript|js)?\n(.*?)```", code, re.DOTALL)
         if fence_match:
             code = fence_match.group(1).strip()
 
         # Rewrite all bare require('pkg') calls to absolute paths so the
-        # script works when run from any /tmp/ directory.  We only rewrite
-        # packages that actually exist in node_modules; unknown packages are
-        # left untouched (they'd fail at runtime but at least not silently).
+        # script works when run from any tmp directory.
         node_modules = REPO_ROOT / "node_modules"
 
         def _rewrite_require(m: re.Match) -> str:
@@ -283,12 +273,11 @@ class SlideGenerator:
             pkg_path = node_modules / pkg
             if pkg_path.exists():
                 return f"require({quote}{pkg_path}{quote})"
-            return m.group(0)  # leave unknown packages as-is
+            return m.group(0)
 
         code = re.sub(r"require\((['\"])([^./][^'\"]*)\1\)", _rewrite_require, code)
 
-        # LLMs sometimes emit the require line twice. Keep only the first
-        # declaration to avoid "Identifier already declared" SyntaxError.
+        # Deduplicate require lines to avoid "Identifier already declared".
         seen: set[str] = set()
         deduped = []
         for line in code.splitlines():
@@ -335,7 +324,7 @@ class SlideGenerator:
             cwd=cwd,
             capture_output=True,
             text=True,
-            timeout=300,  # 5 min hard limit per stage
+            timeout=300,
         )
         if result.returncode != 0:
             raise RuntimeError(

pyproject.toml

@@ -10,7 +10,8 @@ requires-python = ">=3.12"
 
 # Core runtime dependencies (required for the environment to run)
 dependencies = [
-    "google-genai>=1.0.0", # Gemini API client (generator + evaluator + optimizer)
+    "anthropic>=0.40.0",   # Claude API client (generator: Sonnet 4.6, optimizer: Opus 4.6)
+    "google-genai>=1.0.0", # Gemini API client (evaluator: Gemini 3.1 Pro)
     "pydantic>=2.6.0",     # Data models with discriminated unions
     "httpx>=0.27.0",       # HTTP client for client.py
     "loguru>=0.7.0",       # Structured logging for client