Remembering Conversations

Semantic and exact-text search across archived conversations lets you retrieve prior discussions instantly. The Skill indexes your conversation history automatically, enabling you to find past debugging sessions, solutions, and architectural decisions by natural language query or exact text match—like git SHAs or error messages—with optional date or project filters.

Yes, the Skill installs a hook that automatically indexes new conversations in the background, building a permanent archive stored under ~/.config/superpowers. You don't need to manually trigger indexing; conversations are indexed as they occur.

The Skill generates concise AI summaries for each conversation using Claude Haiku with Sonnet fallback, letting you quickly grasp context and decisions without reading full transcripts. Vector indexing with sqlite-vec enables semantic search to find conversations by meaning, not just keywords.

The CLI workflow exposes index, search, verify, repair, and rebuild commands to manage and maintain your archives. Use verify and repair to check archive integrity, and rebuild to regenerate embeddings if needed.

The Skill requires @anthropic-ai/claude-agent-sdk for Claude integration, @xenova/transformers for embedding generation, better-sqlite3 and sqlite-vec for vector indexing and storage. These dependencies handle semantic search, summarization, and persistent archive management automatically.