Store and retrieve structured memories for PACT agents across sessions.
npx skills add https://github.com/ProfSynapse/PACT-Plugin --skill pact-memory
Please help me install this Agent Skill. Skill: pact-memory Source: https://github.com/ProfSynapse/PACT-Plugin/tree/main/pact-plugin/skills/pact-memory Command: npx skills add https://github.com/ProfSynapse/PACT-Plugin --skill pact-memory
PACT Memory provides a persistent, structured memory store for agents to save context, goals, lessons, decisions, and entity references across sessions, enabling cross-session reasoning and faster iteration.
Use pact-memory to save a memory for a session, link relevant files, and enable semantic search when dependencies are installed: /pact-memory save '{"context":"Designing a new authentication flow","goal":"Define token lifecycle","lessons_learned":["Token rotation matters","Handle clock skew"]}'
High-intent search queries and answers about installing and using this skill.
To persist agent context across multiple coding sessions, you can save structured memory objects containing fields like context, goals, lessons learned, and decisions. This enables cross-session reasoning and faster iteration without losing previous work history.
You store and retrieve structured memories for PACT agents by capturing context during Prepare, Architect, Code, and Test phases. Retrieval uses semantic search via embeddings and graph-enhanced retrieval to surface relevant past memories for current tasks.
You do not need vector search dependencies to use memory search functionality. While sqlite-vec and model2vec enable semantic search, the system falls back to functional keyword search if these optional backends are unavailable.
The best way to link relevant files to a saved memory is using the file-linking and memory graph capabilities. This traces context across codebases and projects by connecting stored memory objects directly to associated files.
Semantic search retrieves past agent lessons and entities by using embeddings generated via model2vec and stored in sqlite-vec. This graph-enhanced retrieval surfaces relevant memories based on contextual similarity rather than exact keyword matches.
Yes, you can use an SQLite-based store to save structured memory objects with JSON fields. It captures context, goals, active tasks, decisions, and entity references, providing a persistent local database for agent memory.
arunodaSearch extracted facts locally using Ollama embeddings and sqlite-vec.
AcosmiSearch long-term memory by keyword with semantic retrieval and relevance scores.
pradyumnasagarAutomate inter-skill data transfer and provenance capture via JSON Material Passports.
devCharuzuDesign and deploy production-grade RAG pipelines with chunking, embeddings, vector stores, and evaluation.
