context-manager

You can maintain conversation context by storing compact Markdown logs with title, summary, and status fields in a dedicated directory. This approach compresses conversation state into minimal entries, allowing an assistant to find and reuse prior session context without storing verbose transcripts, directly reducing token usage.

Detect duplicate conversation sessions by scanning log directories using filename matching, fuzzy keyword overlap, or semantic similarity. When a match is found, the system notifies you of the existing session's status and summary, then prompts you to either Ignore, Enhance, or Proceed with the matched conversation.

The best way to structure conversation logs is using Markdown files with specific metadata fields: title, desc, status, last_accessed, summary, and timestamped change bullets. This structured format serves as a single source of truth for session tracking while keeping entries minimal to conserve tokens.

Yes, you can use Markdown files for conversation management by creating and updating them under a dedicated conversations directory. The system performs fuzzy keyword and semantic search on these files to find existing contexts, enabling efficient context reuse across ongoing tasks.

Update session status by managing transitions between in-progress, inconclusive, and completed states within the Markdown log. After each action, prepend concise change bullets to the log, update the last_accessed timestamp, and write a one-line summary to maintain an accurate, minimal record.

Your assistant repeats work because it lacks a centralized conversation log system for duplicate detection. By scanning a conversations directory for exact title matches, fuzzy keywords, or semantic similarity at session start, the assistant can discover and reuse prior context, eliminating duplicated effort.