alma-memory 0.5.0__py3-none-any.whl → 0.7.0__py3-none-any.whl

alma/consolidation/prompts.py

@@ -1,84 +1,84 @@
-"""
-ALMA Consolidation Prompts.
-
-LLM prompts for intelligently merging similar memories.
-"""
-
-# Prompt for merging multiple similar heuristics into one
-MERGE_HEURISTICS_PROMPT = """You are a memory consolidation agent. Given these similar heuristics that have been identified as near-duplicates based on semantic similarity, create a single consolidated heuristic that captures the essence of all.
-
-Similar Heuristics:
-{heuristics}
-
-Create a consolidated heuristic that:
-1. Generalizes the condition to cover all cases
-2. Combines the strategies into a comprehensive approach
-3. Preserves any unique insights from individual heuristics
-
-Output a JSON object with exactly these fields:
-{{
-    "condition": "The generalized condition that triggers this heuristic",
-    "strategy": "The merged strategy combining all approaches",
-    "confidence": <average confidence as a float between 0 and 1>
-}}
-
-Only output the JSON object, no other text."""
-
-# Prompt for merging similar domain knowledge
-MERGE_DOMAIN_KNOWLEDGE_PROMPT = """You are a memory consolidation agent. Given these similar domain knowledge facts that have been identified as near-duplicates, create a single consolidated fact that captures all the information.
-
-Similar Domain Knowledge:
-{knowledge_items}
-
-Create a consolidated fact that:
-1. Combines all unique information
-2. Removes redundancy
-3. Maintains accuracy
-
-Output a JSON object with exactly these fields:
-{{
-    "fact": "The consolidated fact combining all information",
-    "confidence": <average confidence as a float between 0 and 1>
-}}
-
-Only output the JSON object, no other text."""
-
-# Prompt for merging anti-patterns
-MERGE_ANTI_PATTERNS_PROMPT = """You are a memory consolidation agent. Given these similar anti-patterns that have been identified as near-duplicates, create a single consolidated anti-pattern.
-
-Similar Anti-Patterns:
-{anti_patterns}
-
-Create a consolidated anti-pattern that:
-1. Generalizes the pattern description
-2. Combines all reasons why it's bad
-3. Provides a comprehensive alternative
-
-Output a JSON object with exactly these fields:
-{{
-    "pattern": "The generalized pattern to avoid",
-    "why_bad": "Combined explanation of why this pattern is problematic",
-    "better_alternative": "The recommended alternative approach"
-}}
-
-Only output the JSON object, no other text."""
-
-# Prompt for merging outcomes (typically used for summarization rather than true merge)
-MERGE_OUTCOMES_PROMPT = """You are a memory consolidation agent. Given these similar task outcomes, create a summary that captures the key learnings.
-
-Similar Outcomes:
-{outcomes}
-
-Create a summary that:
-1. Identifies the common task type
-2. Notes the overall success/failure pattern
-3. Highlights effective strategies
-
-Output a JSON object with exactly these fields:
-{{
-    "task_type": "The common task type",
-    "summary": "Summary of the outcomes and learnings",
-    "recommended_strategy": "The most effective strategy based on the outcomes"
-}}
-
-Only output the JSON object, no other text."""
+"""
+ALMA Consolidation Prompts.
+
+LLM prompts for intelligently merging similar memories.
+"""
+
+# Prompt for merging multiple similar heuristics into one
+MERGE_HEURISTICS_PROMPT = """You are a memory consolidation agent. Given these similar heuristics that have been identified as near-duplicates based on semantic similarity, create a single consolidated heuristic that captures the essence of all.
+
+Similar Heuristics:
+{heuristics}
+
+Create a consolidated heuristic that:
+1. Generalizes the condition to cover all cases
+2. Combines the strategies into a comprehensive approach
+3. Preserves any unique insights from individual heuristics
+
+Output a JSON object with exactly these fields:
+{{
+    "condition": "The generalized condition that triggers this heuristic",
+    "strategy": "The merged strategy combining all approaches",
+    "confidence": <average confidence as a float between 0 and 1>
+}}
+
+Only output the JSON object, no other text."""
+
+# Prompt for merging similar domain knowledge
+MERGE_DOMAIN_KNOWLEDGE_PROMPT = """You are a memory consolidation agent. Given these similar domain knowledge facts that have been identified as near-duplicates, create a single consolidated fact that captures all the information.
+
+Similar Domain Knowledge:
+{knowledge_items}
+
+Create a consolidated fact that:
+1. Combines all unique information
+2. Removes redundancy
+3. Maintains accuracy
+
+Output a JSON object with exactly these fields:
+{{
+    "fact": "The consolidated fact combining all information",
+    "confidence": <average confidence as a float between 0 and 1>
+}}
+
+Only output the JSON object, no other text."""
+
+# Prompt for merging anti-patterns
+MERGE_ANTI_PATTERNS_PROMPT = """You are a memory consolidation agent. Given these similar anti-patterns that have been identified as near-duplicates, create a single consolidated anti-pattern.
+
+Similar Anti-Patterns:
+{anti_patterns}
+
+Create a consolidated anti-pattern that:
+1. Generalizes the pattern description
+2. Combines all reasons why it's bad
+3. Provides a comprehensive alternative
+
+Output a JSON object with exactly these fields:
+{{
+    "pattern": "The generalized pattern to avoid",
+    "why_bad": "Combined explanation of why this pattern is problematic",
+    "better_alternative": "The recommended alternative approach"
+}}
+
+Only output the JSON object, no other text."""
+
+# Prompt for merging outcomes (typically used for summarization rather than true merge)
+MERGE_OUTCOMES_PROMPT = """You are a memory consolidation agent. Given these similar task outcomes, create a summary that captures the key learnings.
+
+Similar Outcomes:
+{outcomes}
+
+Create a summary that:
+1. Identifies the common task type
+2. Notes the overall success/failure pattern
+3. Highlights effective strategies
+
+Output a JSON object with exactly these fields:
+{{
+    "task_type": "The common task type",
+    "summary": "Summary of the outcomes and learnings",
+    "recommended_strategy": "The most effective strategy based on the outcomes"
+}}
+
+Only output the JSON object, no other text."""