Chroma MCP Server

# Script: `review-and-promote.sh` **DEPRECATION NOTICE:** The `review-and-promote.sh` shell script is deprecated and will be removed in version 0.3.0. Please use the `review-promote` console script installed via the Python package (`chroma-mcp-client review-promote`). This script provides a convenient way to launch the interactive chat history review and promotion workflow. ## Purpose Initiates the `chroma-mcp-client review-and-promote` command within the correct Hatch environment. This command fetches chat entries marked as 'analyzed', displays them to the user, allows searching the codebase for relevant code references, and prompts the user to promote, ignore, or skip each entry. See the main [`chroma-mcp-client` documentation](chroma-mcp-client.md#review-and-promote) for full details on the underlying command and its options. ## Enhanced Features The new version includes several significant enhancements: 1. **Auto-Promote Mode**: - Option to automatically promote entries with high confidence (≥0.8) - Customizable confidence threshold for auto-promotion - Summary of auto-promoted entries at the end of the session - Streamlined workflow for high-confidence entries 2. **Smart Defaults**: - Intelligent defaults for all input fields based on context - Pattern generated based on modification type and content - Best code reference auto-selected based on distance score - Tags suggested from modification type and file extensions - Default confidence from source chat entry 3. **Low Confidence Warnings**: - Visual warnings for entries with confidence < 0.5 - Prompts user to review these entries more carefully - Helps maintain quality in the derived learnings collection 4. **Rich Context Display**: Shows detailed context information for each chat entry, including: - Color-coded confidence scores - Tool sequences with pattern recognition - Diff summaries of code changes - Related code chunks from bidirectional links - Full code context viewing with the 'v' command 5. **Entry Sorting & Filtering**: - Sort by confidence score (highest first) to prioritize valuable interactions - Filter by modification type (refactor, bugfix, feature, etc.) - Set minimum confidence threshold - Calculate and display context richness score (visual indicator) 6. **Improved Promotion Workflow**: - Enhanced bidirectional linking for related code chunks - Auto-selects most relevant code reference - Clear information about code chunks with distance scores - Enhanced description with diff summary - Suggested tags based on modification type and file extensions - Default confidence values from source entry - Option to include/exclude rich context in promoted learnings ## Usage **Recommended**: Use the console script directly: ```bash chroma-mcp-client review-promote [OPTIONS] ``` **Legacy wrapper script (deprecated)**: ```bash ./scripts/review_and_promote.sh [OPTIONS] ``` ```bash # Run with custom day limit ./scripts/review_and_promote.sh --days-limit 3 # Filter by modification type and minimum confidence ./scripts/review_and_promote.sh --modification-type refactor --min-confidence 0.7 # Custom collection names ./scripts/review_and_promote.sh --chat-collection-name my_chats_v2 --learnings-collection-name my_learnings_v2 ``` ## Command Options - `--days-limit N`: How many days back to look for entries (default: 7) - `--fetch-limit N`: Maximum number of entries to fetch (default: 50) - `--chat-collection-name NAME`: Chat history collection name (default: chat_history_v1) - `--learnings-collection-name NAME`: Derived learnings collection name (default: derived_learnings_v1) - `--modification-type TYPE`: Filter by modification type (choices: all, refactor, bugfix, feature, documentation, optimization, test, config, style, unknown; default: all) - `--min-confidence N`: Minimum confidence score threshold (0.0-1.0, default: 0.0) - `--sort-by-confidence`: Sort entries by confidence score (default: true) - `--no-sort-by-confidence`: Don't sort entries by confidence score - `--auto-promote`: Automatically promote high confidence entries without user review - `--auto-promote-threshold N`: Custom confidence threshold for auto-promotion (0.0-1.0, default: 0.8) ## Interactive Commands During the review process, the following commands are available: - `p`: Promote the current entry to derived learnings - `i`: Ignore the entry (marks it as 'ignored') - `s`: Skip the entry (leaves status unchanged) - `v`: View full code context (when available) - `q`: Quit the review process ## Smart Defaults The smart defaults feature means users can often just press Enter to accept reasonable default values for: - Description: Generated from chat entry summary - Pattern: Created based on modification type and entry content - Code reference: Best match auto-selected from bidirectional links or search - Tags: Generated from modification type and file extensions - Confidence: Inherited from source chat entry ## Context Richness Score Each entry displays a context richness indicator (0-5 stars), computed from: - Code context availability - Diff summary presence - Tool sequence information - Related code chunks - Confidence score - Modification type Entries with higher richness scores tend to provide better context for derived learnings. ## Hatch Alias A convenient alias `review-promote` is typically defined in `pyproject.toml`: ```bash # Run via hatch alias with defaults hatch run review-promote # Run via hatch alias with arguments hatch run review-promote --modification-type feature --min-confidence 0.8 # Run with auto-promote enabled hatch run review-promote --auto-promote --auto-promote-threshold 0.75 ``` This script simplifies running the interactive promotion process from the command line.