deepagents-printshop 0.1.0__py3-none-any.whl

tools/pattern_injector.py

@@ -0,0 +1,250 @@
+"""
+Pattern Injector - Milestone 3
+
+Injects learned patterns into agent prompts to improve decision-making.
+Uses historical learnings to guide LLM behavior without hard-coding fixes.
+"""
+
+import json
+from pathlib import Path
+from typing import Dict, Optional
+
+
+class PatternInjector:
+    """
+    Inject learned patterns into agent prompts.
+
+    Provides context-aware prompt augmentation based on historical patterns.
+    """
+
+    def __init__(self, document_type: str = "research_report"):
+        """
+        Initialize pattern injector.
+
+        Args:
+            document_type: Type of document (e.g., 'research_report', 'article', 'technical_doc')
+        """
+        self.document_type = document_type
+        self.patterns_file = Path(".deepagents") / "memories" / document_type / "learned_patterns.json"
+        self.patterns = self._load_patterns()
+
+    def _load_patterns(self) -> Dict:
+        """Load learned patterns from file."""
+        if not self.patterns_file.exists():
+            return {
+                "common_latex_fixes": {},
+                "quality_improvements": [],
+                "recurring_issues": [],
+                "agent_performance": {},
+                "insights": []
+            }
+
+        try:
+            with open(self.patterns_file, 'r', encoding='utf-8') as f:
+                return json.load(f)
+        except Exception as e:
+            print(f"⚠️  Could not load patterns: {e}")
+            return {
+                "common_latex_fixes": {},
+                "quality_improvements": [],
+                "recurring_issues": [],
+                "agent_performance": {},
+                "insights": []
+            }
+
+    def get_context_for_latex_specialist(self) -> str:
+        """
+        Generate prompt context for LaTeX Specialist agent.
+
+        Returns:
+            Formatted context string with historical learnings
+        """
+        if not self.patterns:
+            return ""
+
+        context_parts = []
+
+        # Add common fixes
+        if self.patterns.get("common_latex_fixes"):
+            context_parts.append("## Historical Patterns - Common LaTeX Issues\n")
+            context_parts.append("Based on analysis of previous documents, the following issues appear frequently:\n")
+
+            # Sort by frequency
+            fixes = sorted(
+                self.patterns["common_latex_fixes"].items(),
+                key=lambda x: x[1]["count"],
+                reverse=True
+            )
+
+            for fix, data in fixes[:10]:  # Top 10
+                context_parts.append(f"- **{fix}** (seen {data['count']}x)")
+
+            context_parts.append("\n💡 Consider checking for these issues proactively.\n")
+
+        # Add recurring recommendations
+        if self.patterns.get("recurring_issues"):
+            context_parts.append("\n## Recurring Recommendations\n")
+            context_parts.append("These suggestions have been made across multiple documents:\n")
+
+            for issue in self.patterns["recurring_issues"][:5]:  # Top 5
+                context_parts.append(f"- {issue}")
+
+            context_parts.append("\n💡 Pay special attention to these areas.\n")
+
+        # Add quality baseline
+        if self.patterns.get("quality_improvements"):
+            scores = [imp["score"] for imp in self.patterns["quality_improvements"]]
+            if scores:
+                avg_score = sum(scores) / len(scores)
+                max_score = max(scores)
+
+                context_parts.append(f"\n## Quality Baseline\n")
+                context_parts.append(f"- Average historical quality score: {avg_score:.1f}/100\n")
+                context_parts.append(f"- Best score achieved: {max_score}/100\n")
+                context_parts.append(f"- Target for this document: {min(max_score + 5, 100)}/100\n")
+
+        return "\n".join(context_parts)
+
+    def get_context_for_content_editor(self) -> str:
+        """
+        Generate prompt context for Content Editor agent.
+
+        Returns:
+            Formatted context string with historical learnings
+        """
+        if not self.patterns:
+            return ""
+
+        context_parts = []
+
+        # Add recurring content issues (if we track them in the future)
+        if self.patterns.get("recurring_issues"):
+            # Filter for content-related issues
+            content_issues = [
+                issue for issue in self.patterns["recurring_issues"]
+                if any(keyword in issue.lower() for keyword in ["readability", "grammar", "style", "clarity"])
+            ]
+
+            if content_issues:
+                context_parts.append("## Historical Patterns - Content Issues\n")
+                context_parts.append("Based on previous content reviews:\n")
+
+                for issue in content_issues[:5]:
+                    context_parts.append(f"- {issue}")
+
+                context_parts.append("\n💡 Focus on these areas during review.\n")
+
+        # Add quality expectations
+        if self.patterns.get("quality_improvements"):
+            scores = [imp["score"] for imp in self.patterns["quality_improvements"]]
+            if scores:
+                avg_score = sum(scores) / len(scores)
+                context_parts.append(f"\n## Quality Expectations\n")
+                context_parts.append(f"Previous documents achieved an average quality of {avg_score:.1f}/100.\n")
+                context_parts.append(f"Aim for content that will support or exceed this quality level.\n")
+
+        return "\n".join(context_parts)
+
+    def get_context_for_visual_qa(self) -> str:
+        """
+        Generate prompt context for Visual QA agent.
+
+        Returns:
+            Formatted context string with historical learnings
+        """
+        if not self.patterns:
+            return ""
+
+        context_parts = []
+
+        # Add visual-specific recurring issues
+        if self.patterns.get("recurring_issues"):
+            visual_issues = [
+                issue for issue in self.patterns["recurring_issues"]
+                if any(keyword in issue.lower() for keyword in ["typography", "spacing", "layout", "formatting"])
+            ]
+
+            if visual_issues:
+                context_parts.append("## Historical Patterns - Visual Issues\n")
+                context_parts.append("These visual/formatting issues have been identified before:\n")
+
+                for issue in visual_issues[:5]:
+                    context_parts.append(f"- {issue}")
+
+                context_parts.append("\n💡 Check carefully for these in the PDF analysis.\n")
+
+        return "\n".join(context_parts)
+
+    def get_context_for_author(self) -> str:
+        """
+        Generate prompt context for Author/Research agent.
+
+        Returns:
+            Formatted context string with historical learnings
+        """
+        if not self.patterns:
+            return ""
+
+        context_parts = []
+
+        # Add general quality insights
+        if self.patterns.get("insights"):
+            context_parts.append("## Document Generation Insights\n")
+
+            for insight in self.patterns["insights"][:3]:
+                context_parts.append(f"**{insight['title']}:**")
+                context_parts.append(f"{insight['description']}")
+                context_parts.append(f"💡 {insight['recommendation']}\n")
+
+        return "\n".join(context_parts)
+
+    def get_summary(self) -> str:
+        """
+        Get a brief summary of available patterns.
+
+        Returns:
+            Summary string
+        """
+        if not self.patterns or not self.patterns_file.exists():
+            return f"No learned patterns available yet for document type: {self.document_type}"
+
+        metadata = self.patterns.get("metadata", {})
+        num_fixes = len(self.patterns.get("common_latex_fixes", {}))
+        num_issues = len(self.patterns.get("recurring_issues", []))
+        num_insights = len(self.patterns.get("insights", []))
+
+        return (
+            f"Learned patterns for '{self.document_type}' from {metadata.get('documents_analyzed', 0)} documents:\n"
+            f"  - {num_fixes} common LaTeX fixes\n"
+            f"  - {num_issues} recurring recommendations\n"
+            f"  - {num_insights} actionable insights"
+        )
+
+
+def demo():
+    """Demonstrate pattern injection."""
+    print("\n" + "=" * 60)
+    print("🧠 Pattern Injector - Demonstration")
+    print("=" * 60)
+
+    injector = PatternInjector()
+
+    print("\n" + injector.get_summary())
+
+    print("\n" + "=" * 60)
+    print("📝 Context for LaTeX Specialist:")
+    print("=" * 60)
+    print(injector.get_context_for_latex_specialist())
+
+    print("\n" + "=" * 60)
+    print("👁️ Context for Visual QA:")
+    print("=" * 60)
+    print(injector.get_context_for_visual_qa())
+
+    print("\n" + "=" * 60)
+    print("✅ Pattern injection ready!")
+    print("=" * 60)
+
+
+if __name__ == "__main__":
+    demo()