Merge pull request #125 from vicsanity623/vicsanity623-patch-1

Vicsanity623 patch 1

Author: vicsanity623

src/pyob/entrance.py

@@ -492,22 +492,42 @@ def build_initial_analysis(self):
             proj_prompt, lambda t: len(t) > 5, context="Project Genesis"
         ).strip()
         content = f"# Project Analysis\n\n**Project Summary:**\n{project_summary}\n\n---\n\n## 📂 File Directory\n\n"
+        file_structures = {}
         for f_path in all_files:
             rel = os.path.relpath(f_path, self.target_dir)
-            logger.info(f"Deep Symbolic Parsing: {rel}")
             with open(f_path, "r", encoding="utf-8", errors="ignore") as f:
                 full_code = f.read()
             self.update_ledger_for_file(rel, full_code)
-            structure_dropdowns = self.code_parser.generate_structure_dropdowns(
+            file_structures[rel] = self.code_parser.generate_structure_dropdowns(
                 f_path, full_code
             )
-            sum_prompt = f"Provide a one-sentence plain text summary of what the file `{rel}` does. \n\nCRITICAL: Do NOT include any HTML tags, <details> blocks, or code signatures in your response. Just the sentence.\n\nFile Structure for context:\n{structure_dropdowns}"
-            desc = self.llm_engine.get_valid_llm_response(
-                sum_prompt, lambda t: "<details>" not in t and len(t) > 5, context=rel
-            ).strip()
-            content += (
-                f"### `{rel}`\n**Summary:** {desc}\n\n{structure_dropdowns}\n---\n"
-            )
+
+        logger.info(f"Batching Deep Symbolic Parsing for {len(all_files)} files...")
+        batch_prompt = "Provide a succinct one-sentence plain text summary for EACH of the following files based on their structure. CRITICAL: Output MUST be strictly in 'filepath: summary' format on each line. Do NOT include any HTML tags, markdown, <details> blocks, or code signatures.\n\n"
+        for rel_path, struct in file_structures.items():
+            batch_prompt += f"File `{rel_path}` Structure:\n{struct}\n\n"
+
+        batch_response = self.llm_engine.get_valid_llm_response(
+            batch_prompt, lambda t: ":" in t and len(t) > 10, context="Batch Genesis"
+        ).strip()
+
+        summaries = {}
+        for line in batch_response.splitlines():
+            if ":" in line:
+                path_part, summary_part = line.split(":", 1)
+                path_clean = path_part.replace("`", "").replace("*", "").strip()
+                summaries[path_clean] = summary_part.strip()
+
+        for f_path in all_files:
+            rel = os.path.relpath(f_path, self.target_dir)
+            structured_summary = summaries.get(rel, "No summary generated.")
+            if structured_summary == "No summary generated.":
+                for k, v in summaries.items():
+                    if k in rel or rel in k:
+                        structured_summary = v
+                        break
+            content += f"### `{rel}`\n**Summary:** {structured_summary}\n\n{file_structures[rel]}\n---\n"
+
         with open(self.analysis_path, "w", encoding="utf-8") as f:
             f.write(content)
         self.save_ledger()

src/pyob/get_valid_edit.py

@@ -90,7 +90,9 @@ def _fetch_llm_with_retries(
         self, prompt: str, display_name: str, attempts: int
     ) -> tuple[str, int]:
         is_cloud = (
-            os.environ.get("GITHUB_ACTIONS") == "true" or os.environ.get("CI") == "true"
+            os.environ.get("GITHUB_ACTIONS") == "true"
+            or os.environ.get("CI") == "true"
+            or "GITHUB_RUN_ID" in os.environ
         )
         key_cooldowns = getattr(self, "key_cooldowns", {})
 
@@ -128,15 +130,15 @@ def _fetch_llm_with_retries(
                     prompt, key=None, context=display_name, gh_model=gh_model
                 )
             else:
-                logger.info("\n[Attempting Local Ollama]")
+                logger.info("\n[All keys exhausted. Falling back to Local Ollama]")
                 response = getattr(self, "_stream_single_llm")(
                     prompt, key=None, context=display_name
                 )
 
             if response.startswith("ERROR_CODE_429"):
                 if key:
-                    key_cooldowns[key] = time.time() + 60
-                    logger.warning("Key rate limited. Rotating instantly...")
+                    key_cooldowns[key] = time.time() + 120
+                    logger.warning("Key rate limited. Pivoting to next key...")
                     attempts += 1
                     continue
                 elif "RateLimitReached" in response:
@@ -165,6 +167,17 @@ def _fetch_llm_with_retries(
                 continue
 
             if response.startswith("ERROR_CODE_") or not response.strip():
+                if key and "429" not in (response or ""):
+                    key_cooldowns[key] = time.time() + 10
+
+                if available_keys:
+                    logger.warning(
+                        f"Engine failed with error: {str(response)[:60]}... Rotating..."
+                    )
+                    attempts += 1
+                    time.sleep(2)
+                    continue
+
                 logger.warning("API Error or Empty Response. Sleeping 60s...")
                 time.sleep(60)
                 attempts += 1

src/pyob/models.py

@@ -259,19 +259,18 @@ def get_valid_llm_response_engine(
         or os.environ.get("CI") == "true"
         or "GITHUB_RUN_ID" in os.environ
     )
-    all_keys = list(key_cooldowns.keys())
-
     while True:
         key = None
         now = time.time()
         # available_keys are Gemini keys
-        available_keys = [k for k in all_keys if now > key_cooldowns[k]]
+        gemini_keys = [k for k in list(key_cooldowns.keys()) if "github" not in k]
+        available_keys = [k for k in gemini_keys if now > key_cooldowns[k]]
         response_text = None
 
         if available_keys:
             key = available_keys[attempts % len(available_keys)]
             logger.info(
-                f"Attempting Gemini Key {attempts % len(available_keys) + 1}/{len(available_keys)}"
+                f"Attempting Gemini Key {attempts % len(available_keys) + 1}/{len(gemini_keys)}"
             )
             response_text = stream_single_llm(prompt, key=key, context=context)
         elif is_cloud:
@@ -290,15 +289,18 @@ def get_valid_llm_response_engine(
                     prompt, key=None, context=context, gh_model="Llama-3"
                 )
         else:
-            logger.info(" Using Local Ollama Engine...")
+            logger.info(
+                " All Gemini keys exhausted. Falling back to Local Ollama Engine..."
+            )
             response_text = stream_single_llm(prompt, key=None, context=context)
 
         # --- ERROR HANDLING BLOCK ---
         if not response_text or response_text.startswith("ERROR_CODE_"):
             # 1. Handle Gemini 429 (Minute limits)
             if key and response_text and "429" in response_text:
-                key_cooldowns[key] = time.time() + 60
-                logger.warning(f"Key {key[-4:]} rate-limited. Rotating...")
+                key_cooldowns[key] = time.time() + 120
+                logger.warning(f"Key {key[-4:]} rate-limited. Pivoting to next key...")
+                attempts += 1
                 # Immediate pivot attempt for this loop
                 if is_cloud:
                     logger.warning(
@@ -307,6 +309,8 @@ def get_valid_llm_response_engine(
                     response_text = stream_single_llm(
                         prompt, key=None, context=context, gh_model="Llama-3"
                     )
+                else:
+                    continue
 
             # 2. Handle GitHub 429 (Daily Quota Limits)
             if (
@@ -345,8 +349,21 @@ def get_valid_llm_response_engine(
                         prompt, key=None, context=context, gh_model="Phi-4"
                     )
 
-            # 4. Final Fail-Safe Sleep
+            # 4. Final Catch-All / Fail-Safe Sleep
             if not response_text or response_text.startswith("ERROR_CODE_"):
+                if key and "429" not in (response_text or ""):
+                    key_cooldowns[key] = (
+                        time.time() + 10
+                    )  # Short cooldown for unknown errors prevent tight loops
+
+                if available_keys:
+                    logger.warning(
+                        f"Engine failed with error: {str(response_text)[:60]}... Rotating..."
+                    )
+                    attempts += 1
+                    time.sleep(2)
+                    continue
+
                 wait = 90
                 logger.warning(
                     f"All Engines failed or exhausted. Sleeping {wait}s for refill..."