feedback history and prompt work better

AGENTS.md

@@ -98,3 +98,66 @@ Successfully implemented a new menu item and functionality for generating theme
 The implementation maintains backward compatibility while adding valuable feedback functionality to improve prompt generation quality over time.
 
 Too many tests, so I moved all of them into the tests directory.
+
+---
+
+# Task: Implement feedback_historic.json cyclic buffer system (30 items)
+
+## Changes Made
+
+### 1. Added Feedback Historic System
+- Created `feedback_historic.json` file to store previous feedback words (without weights)
+- Implemented a cyclic buffer system with 30-item capacity (feedback00-feedback29)
+- When new feedback is generated (6 words), they become feedback00-feedback05
+- All existing items shift down by 6 positions
+- Items beyond feedback29 are discarded
+
+### 2. Updated Class Initialization
+- Added `feedback_historic` attribute to `JournalPromptGenerator` class
+- Updated `__init__` method to load `feedback_historic.json`
+- Added `_load_feedback_historic()` method to load historic feedback words
+- Added `_save_feedback_historic()` method to save historic feedback words (keeping only first 30)
+
+### 3. Enhanced Feedback Words Management
+- Updated `add_feedback_words_to_history()` method to:
+  - Extract just the words from current feedback words (no weights)
+  - Add 6 new words to the historic buffer
+  - Shift all existing words down by 6 positions
+  - Maintain 30-item limit by discarding oldest items
+- Updated `update_feedback_words()` to automatically call `add_feedback_words_to_history()`
+
+### 4. Improved AI Prompt Generation
+- Updated `generate_theme_feedback_words()` method to include historic feedback words in API call
+- The prompt now includes three sections:
+  1. Previous prompts (historic prompts)
+  2. Current feedback themes (with weights)
+  3. Historic feedback themes (just words, no weights)
+- This helps the AI avoid repeating previously used theme words
+
+### 5. Data Structure Design
+- Historic feedback words are stored as a list of dictionaries with keys (feedback00, feedback01, etc.)
+- Each dictionary contains only the word (no weight field)
+- Structure mirrors `prompts_historic.json` but for feedback words
+- 30-item limit provides sufficient history while preventing excessive repetition
+
+## Testing
+- Created comprehensive test to verify cyclic buffer functionality
+- Tested that new items are added at the beginning (feedback00-feedback05)
+- Verified that existing items shift down correctly
+- Confirmed 30-item limit is enforced (oldest items are dropped)
+- Tested that historic feedback words are included in AI prompts
+- Verified that weights are not stored in historic buffer (only words)
+
+## Result
+Successfully implemented a feedback historic cyclic buffer system that:
+1. Stores previous feedback words in `feedback_historic.json` (30-item limit)
+2. Automatically adds new feedback words to history when they are updated
+3. Includes historic feedback words in AI prompts to avoid repetition
+4. Maintains consistent data structure with the rest of the system
+5. Provides a memory of previous theme words to improve AI suggestions over time
+
+The system now has a complete feedback loop where:
+- Historic prompts and feedback words inform new theme word generation
+- New theme words are rated by users and become current feedback words
+- Current feedback words are added to the historic buffer
+- Historic feedback words help avoid repetition in future theme word generation