key snippet gc

ra_aid/__main__.py

@@ -610,9 +610,6 @@ def main():
                     memory=planning_memory,
                     config=config,
                 )
-                
-            # Run cleanup tasks before exiting database context
-            run_cleanup()
 
     except (KeyboardInterrupt, AgentInterrupt):
         print()
@@ -620,18 +617,5 @@ def main():
         print()
         sys.exit(0)
 
-
-def run_cleanup():
-    """Run cleanup tasks after main execution."""
-    try:
-        # Import the key facts cleaner agent
-        from ra_aid.agents.key_facts_gc_agent import run_key_facts_gc_agent
-        
-        # Run the key facts garbage collection agent regardless of the number of facts
-        run_key_facts_gc_agent()
-    except Exception as e:
-        logger.error(f"Failed to run cleanup tasks: {str(e)}")
-
-
 if __name__ == "__main__":
     main()

ra_aid/agents/__init__.py

@@ -1,26 +1,18 @@
 """
-Key facts gc agent package.
+Agent package for various specialized agents.
 
-This package contains the agent responsible for cleaning up key facts 
+This package contains agents responsible for specific tasks such as
-in the database when they exceed a certain threshold.
+cleaning up key facts and key snippets in the database when they
+exceed certain thresholds.
+
+Includes agents for:
+- Key facts garbage collection
+- Key snippets garbage collection
 """
 
 from typing import Optional
 
-def run_key_facts_gc_agent(max_facts: int = 20) -> None:
+from ra_aid.agents.key_facts_gc_agent import run_key_facts_gc_agent
-    """
+from ra_aid.agents.key_snippets_gc_agent import run_key_snippets_gc_agent
-    Run the key facts gc agent to reduce key facts to the specified maximum.
+
-    
+__all__ = ["run_key_facts_gc_agent", "run_key_snippets_gc_agent"]
-    This agent evaluates the importance of key facts and removes the least important ones
-    when the total count exceeds the maximum threshold.
-    
-    Args:
-        max_facts: Maximum number of key facts to keep (defaults to 20)
-    """
-    # This is a placeholder function that will be implemented later
-    # The actual implementation will:
-    # 1. Fetch all key facts from the database
-    # 2. Evaluate their importance based on certain criteria
-    # 3. Sort them by importance
-    # 4. Delete the least important ones until only max_facts remain
-    pass

ra_aid/agents/key_snippets_gc_agent.py

@@ -0,0 +1,85 @@
+"""
+Key snippets gc agent implementation.
+
+This agent is responsible for maintaining the code snippet knowledge base by pruning less important
+snippets when the total number exceeds a specified threshold. The agent evaluates all
+key snippets and deletes the least valuable ones to keep the database clean and relevant.
+"""
+
+from typing import List
+
+from langchain_core.tools import tool
+from rich.console import Console
+from rich.markdown import Markdown
+from rich.panel import Panel
+
+from ra_aid.agent_utils import create_agent, run_agent_with_retry
+from ra_aid.database.repositories.key_snippet_repository import KeySnippetRepository
+from ra_aid.llm import initialize_llm
+from ra_aid.prompts.key_snippets_gc_prompts import KEY_SNIPPETS_GC_PROMPT
+from ra_aid.tools.memory import delete_key_snippets, log_work_event, _global_memory
+
+
+console = Console()
+key_snippet_repository = KeySnippetRepository()
+
+
+def run_key_snippets_gc_agent() -> None:
+    """Run the key snippets gc agent to maintain a reasonable number of key snippets.
+    
+    The agent analyzes all key snippets and determines which are the least valuable,
+    deleting them to maintain a manageable collection size of high-value snippets.
+    """
+    # Get the count of key snippets
+    snippets = key_snippet_repository.get_all()
+    snippet_count = len(snippets)
+    
+    # Display status panel with snippet count included
+    console.print(Panel(f"Gathering my thoughts...\nCurrent number of key snippets: {snippet_count}", title="🗑 Garbage Collection"))
+    
+    # Only run the agent if we actually have snippets to clean
+    if snippet_count > 0:
+        # Get all snippets as a formatted string for the prompt
+        snippets_dict = key_snippet_repository.get_snippets_dict()
+        formatted_snippets = "\n".join([
+            f"Snippet #{k}: filepath={v['filepath']}, line_number={v['line_number']}, description={v['description']}\n```python\n{v['snippet']}\n```" 
+            for k, v in snippets_dict.items()
+        ])
+        
+        # Retrieve configuration
+        llm_config = _global_memory.get("config", {})
+
+        # Initialize the LLM model
+        model = initialize_llm(
+            llm_config.get("provider", "anthropic"),
+            llm_config.get("model", "claude-3-7-sonnet-20250219"),
+            temperature=llm_config.get("temperature")
+        )
+        
+        # Create the agent with the delete_key_snippets tool
+        agent = create_agent(model, [delete_key_snippets])
+        
+        # Format the prompt with the current snippets
+        prompt = KEY_SNIPPETS_GC_PROMPT.format(key_snippets=formatted_snippets)
+        
+        # Set up the agent configuration
+        agent_config = {
+            "recursion_limit": 50  # Set a reasonable recursion limit
+        }
+        
+        # Run the agent
+        run_agent_with_retry(agent, prompt, agent_config)
+        
+        # Get updated count
+        updated_snippets = key_snippet_repository.get_all()
+        updated_count = len(updated_snippets)
+        
+        # Show info panel with updated count
+        console.print(
+            Panel(
+                f"Cleaned key snippets: {snippet_count} → {updated_count}",
+                title="🗑 GC Complete"
+            )
+        )
+    else:
+        console.print(Panel("No key snippets to clean.", title="🗑 GC Info"))

ra_aid/prompts/key_snippets_gc_prompts.py

@@ -0,0 +1,57 @@
+"""
+Key snippets gc-specific prompts for the AI agent system.
+
+This module contains the prompt for the key snippets gc agent that is
+responsible for evaluating and trimming down the stored key snippets to keep
+only the most valuable ones, ensuring that the collection remains manageable.
+"""
+
+KEY_SNIPPETS_GC_PROMPT = """
+You are a Key Snippets Cleaner agent responsible for maintaining the code snippet knowledge base by pruning less important snippets.
+
+<key snippets>
+{key_snippets}
+</key snippets>
+
+Task:
+Your task is to analyze all the key snippets in the system and determine which ones should be kept and which ones should be removed.
+
+Guidelines for evaluation:
+1. Review all key snippets and their IDs
+2. Identify which snippets are lowest value/most ephemeral based on:
+   - Relevance to the overall project
+   - Specificity and actionability of the code
+   - Long-term value vs. temporary relevance
+   - Uniqueness of the information (avoid redundancy)
+   - How fundamental the snippet is to understanding the codebase
+
+3. Trim down the collection to keep no more than 10 highest value, longest-lasting snippets
+4. For each snippet you decide to delete, provide a brief explanation of your reasoning
+
+Retention priority (from highest to lowest):
+- Core architectural code that demonstrates project structure
+- Critical implementation details that affect multiple parts of the system
+- Important design patterns and conventions
+- API endpoints and interfaces
+- Configuration requirements
+- Complex algorithms
+- Error handling patterns
+- Testing approaches
+- Simple helper functions or boilerplate code that is easily rediscovered
+- If there are contradictory snippets, that probably means that the older snippet is outdated and should be deleted.
+
+For snippets of similar importance, prefer to keep more recent snippets if they supersede older information.
+
+Output:
+1. List the IDs of snippets to be deleted using the delete_key_snippets tool with the IDs provided as a list [ids...], NOT as a comma-separated string
+2. Provide a brief explanation for each deletion decision
+3. Explain your overall approach to selecting which snippets to keep
+
+IMPORTANT:
+- Use the delete_key_snippets tool with multiple IDs at once in a single call, rather than making multiple individual deletion calls
+- The delete_key_snippets tool accepts a list of IDs in the format [id1, id2, id3, ...], not as a comma-separated string
+- Batch deletion is much more efficient than calling the deletion function multiple times
+- Collect all IDs to delete first, then make a single call to delete_key_snippets with the complete list
+
+Remember: Your goal is to maintain a concise, high-value code snippet collection that preserves essential project understanding while removing ephemeral or less critical snippets.
+"""

ra_aid/tools/memory.py

@@ -242,6 +242,17 @@ def emit_key_snippet(snippet_info: SnippetInfo) -> str:
     )
 
     log_work_event(f"Stored code snippet #{snippet_id}.")
+    
+    # Check if we need to clean up snippets (more than 20)
+    all_snippets = key_snippet_repository.get_all()
+    if len(all_snippets) > 20:
+        # Trigger the key snippets cleaner agent
+        try:
+            from ra_aid.agents.key_snippets_gc_agent import run_key_snippets_gc_agent
+            run_key_snippets_gc_agent()
+        except Exception as e:
+            logger.error(f"Failed to run key snippets cleaner: {str(e)}")
+    
     return f"Snippet #{snippet_id} stored."