.agents/memory/MEMORY.md

Current Context

The current focus is implementing the Knowledge Architecture (KA-1) and Procedural Memory P1 sprint deliverables within the signetai monorepo. Work is concentrated on the packages/core module to establish the foundational database schema and graph relationships for "skills as first-class graph nodes," followed by the dashboard implementation.

Active Projects

Procedural Memory P1 (SPRINT-BRIEF.md) Location: packages/core/src (Types, Migrations), packages/dashboard Status: In Progress. Recently completed configuration and migration infrastructure. Next steps involve the skill entity creation and graph relation implementation. Current Blockers: None. Next Steps: Execute Deliverable 2 & 3 together (Migration of skill_meta details and Skill entity creation). Create skills.ts in the core package to handle the filesystem registry. Implement the Skill Install Flow (Deliverable 6). Develop the Query API for skill retrieval (Deliverable 5). Technical Detail: Must use Migration 018 for the skill_meta table (added to migrations index). The skill_meta table will include entity_id, agent_id, and specific metadata columns.

Knowledge Architecture (SPRINT-BRIEF-KA1.md) Location: packages/core/src (Types, Migrations, Knowledge Graph) Status: Nearly Complete. Migration, types, and helper modules are drafted. Next Steps: Finalize the registration of Migration 019 in the migrations index. Verify that Entity interface consumers in graph-transactions.ts and skill-graph.ts correctly handle the new agent_id requirements. Complete the knowledge-graph.ts read/write helpers module. Technical Detail: Migration 019 creates the knowledge_assets table. The knowledge-graph.ts module must expose core types from the package index.

Session Continuity Strategy Location: Architectural discussion / Documentation Status: Planning Phase. Next Steps: Clarify the interaction between session continuity protocols and existing database compaction flows to ensure non-breaking integration.

Recent Work

Procedural Memory Configuration & Migration: Added PipelineProceduralConfig interface to packages/core/src/types.ts. Updated PipelineV2Config to include the procedural field. Wired defaults in packages/core/src/memory-config.ts with full YAML parsing support. Created Migration 018 in packages/core/src/migrations/ for the skill_meta table (schema includes entity_id, agent_id, created_at, etc.). Registered Migration 018 in packages/core/src/migrations/index.ts. Knowledge Architecture Implementation: Created Migration 019 for the knowledge_assets table. Added KA types and constants to packages/core/src/types.ts. Updated the Entity interface to include the new KA types. Created the knowledge-graph.ts module with read/write helpers. Updated production INSERT locations in graph-transactions.ts and skill-graph.ts to include agent_id. Marketing & Documentation: Combined and archived email files. Reframed marketing copy for the "You Think Signet Is a Memory System" initiative. Migrated VISION.md content to the "It Learns Now" blog post.

Technical Notes

Repository Structure: Working in /home/nicholai/signet/signetai on the main branch. Database: SQLite backend. UI Stack: packages/dashboard utilizes Svelte and TypeScript with Lucide icons. Migration Protocol: Before writing any new migration, strictly check packages/core/src/migrations/index.ts for the next available number to prevent collisions. File System: The skill_meta table requires a Foreign Key relationship to assets (entity_id) and includes a column for agent_id. Package Exports: The knowledge-graph.ts module must export the core types in packages/core/index.ts for the dashboard to access them. Known Issues: There is an ongoing issue with Daemon startup crashes related to the 1Password SDK WASM dependency that requires debugging.

Rules & Warnings

No Production DB Deletions: Never run deletion commands on the production database without explicit confirmation. Remote Server Access: Never restart remote machines or servers without permission. UI Implementation: Opus must perform all UI code changes manually; never delegate UI work to subagents. Agent Identity: I am Claude, a large language model. I should be addressed as "he" or by name in correspondence.