sdcorejs-memories

To persist project conventions across Claude Code agent sessions, this Skill writes durable memories to a local project directory, loading frontmatter at session start and full content on demand. It also mirrors memory bytes directly into Claude Code's local per-project directory for automatic surfacing.

Cross-session memory for agent workflows is durable storage that saves lasting facts like stakeholder constraints and recurring anti-patterns. You need it when onboarding repeats across coding sessions and forgotten conventions cause future agents to disregard established project rules.

To save durable knowledge for NestJS or NextJS workflows, trigger a write by explicitly requesting it or making a durable correction. The agent stores per-track memory files under a dedicated memories directory, organizing knowledge by the specific workflow track.

Yes, this memory Skill requires a Git repository to function correctly. It resolves the target project root using Git, which establishes the base directory path needed to read and write memory files within the project structure.

The agent handles duplicate topics by applying slug and index guidance during the memory write process. This duplicate detection mechanism ensures that recurring conventions are updated accurately rather than creating redundant memory files.

Limitations of file-based SDLC memory include best-effort mirroring into Claude Code's directory, which is not guaranteed outside that environment. It also requires explicit user triggers or detected durable corrections to write, meaning transient session details are not captured automatically.