What problem does it solve?
Prevents “probably/likely” assumptions and incorrect guidance in large codebases by forcing the agent to retrieve and cite relevant, source-of-truth memories before taking consequential actions.
Core Features & Use Cases
- Contextual memory retrieval using scoped queries by expected file paths and command context to find the most relevant prior guidance.
- Metadata narrowing and candidate inspection to efficiently shortlist memories, then inspect full details for accuracy.
- Trust and staleness discipline by prioritizing severity/weight/scope specificity and highlighting verification state so conflicting cues are escalated instead of averaged.
- Output citation requirement ensuring responses and plans cite memory IDs and their linked sources (and re-retrieve with tighter parameters if citations are missing).
Quick Start
Use retrieving-memory before making changes to a subsystem you have not touched this run by running a scoped memory query for the file paths you expect to edit and the command pipeline you plan to execute.
