Optional: local embeddings (offline, ~2GB for PyTorch)

1---
2name: memora
3description: Persistent memory and knowledge management. Use at session start to load context, when the user asks about past work or stored knowledge, and when saving important information for future sessions.
4user-invocable: false
5---
6 
7## Session Start
8 
9At the beginning of each session, proactively search memories for context related to the current project or working directory using `memory_hybrid_search`. Briefly summarize relevant findings to establish context.
10 
11## Memory Search
12 
13When the user asks about past work, stored knowledge, or previously discussed topics:
14 
151. Use `memory_hybrid_search` to find relevant memories
162. Use `memory_semantic_search` for pure meaning-based lookup
173. Summarize findings and cite memory IDs (e.g., "Memory #42 shows...")
18 
19For research/recall questions, always search memories first before answering.
20 
21## Saving Memories
22 
23When the user asks to remember something, or when important decisions, patterns, or context emerge during a session:
24 
251. Use `memory_create` for general knowledge and notes
262. Use `memory_create_todo` for tasks and action items
273. Use `memory_create_issue` for bugs and problems
284. Use appropriate tags to organize memories
29 
30## Avoiding Duplicates
31 
32Before creating a new memory, use `memory_hybrid_search` to check if similar content already exists. If a near-match is found, use `memory_update` to extend the existing memory instead of creating a duplicate.
33 
34## Linking Related Memories
35 
36When creating or finding memories that relate to each other, use `memory_link` to establish explicit connections. This strengthens the knowledge graph and improves future recall.
37