Memory Manager V3.9.0

Problem Statement: The Hidden Token Drain in AI Sessions

In prolonged conversations, AI assistants automatically load all memory files from the .workbuddy/memory directory into the context window. Over time, these files accumulate: obsolete dialogues, duplicated memory fragments, and stale cache data. The direct consequences are soaring initial token consumption per session, leading to slower responses and increased costs, while low-quality memories can distort AI judgment.

Specific symptoms include:
- Token Budget Overrun: An inflated memory file can consume thousands of tokens, crowding out space for new tasks.
- Inefficient Search: Retrieving specific information from a chaotic memory mass forces the AI to process excessive irrelevant content.
- Cache Liability: Piled-up temporary data occupies disk space and may degrade performance.

Core Functionality: Precision Trimming of Memory Streams

Memory Manager V3.9.0, as an AI memory accelerator, achieves token throttling through an automated toolchain. Its workflow revolves around three key steps:

1. Memory Analysis and Quantification
   - Commands like analyze or token-check scan the memory directory and generate a token usage report.
   - token-trends provides historical consumption trends to identify abnormal spikes.
   - The rank command prioritizes memories by importance, offering a basis for cleanup.

2. Intelligent Cleanup and Optimization
   - Deduplication and Cleaning: dedup detects duplicate memories, while auto-clean performs differentiated cleaning (e.g., removing old caches, compressing outdated data).
   - Budget Control: The throttle command compresses memory before injection to ensure the total token count stays within a set budget.
   - Precision Loading: load --days N or load --query [topic] loads only essential memories, avoiding full scans.

3. Safety and Operational Safeguards
   - Three-tier operation permissions: read-only actions execute directly; cleanup actions default to preview (dry-run) and require user confirmation; destructive actions must explicitly use --execute.
   - Health Checks: The doctor command diagnoses system status, and cache-reminder monitors cache accumulation.
   - Backup and Recovery: export and import support versioned memory management.

Scope and Limitations

This skill is designed exclusively for the .workbuddy/memory directory. It does not apply to:
- General file management (outside the memory directory)
- Code debugging or refactoring (requires dedicated code tools)
- System-level disk cleanup (involves the global file system)

Key constraints:
- Encoding Dependency: Memory files must be UTF-8; otherwise, UnicodeDecodeError may occur.
- Permission Requirements: Write access to the .workbuddy/ directory is mandatory; otherwise, PermissionError is triggered.
- Feature Fallback: Optional dependencies (e.g., tiktoken for precise token calculation) cause automatic degradation when missing, but core cleanup remains functional.

Usage Tip: After the first run, token data accumulates gradually. It is recommended to regularly execute token-trends to monitor trends and proactively control memory injection scale with throttle.