Project memory is useful until it starts carrying old assumptions, duplicate rules, and subtle contradictions.
OpenCode Plugin Memory v0.2.0 adds a maintenance layer without turning memory into an opaque database or an autonomous rewrite pipeline.
The release introduces:
- atomic superseding: add the replacement memory and archive the old one together;
- archive metadata that records why an entry was retired and what replaced it;
- a restricted background audit that can identify duplicate, refinement, and contradiction candidates;
- memory_review, including on-demand refresh and dismissals that expire when the underlying memories change. The important boundary: the background audit only produces findings in review.json. It never modifies MEMORY.md. The main session still decides whether a memory should be updated, superseded, or archived.
Memory remains project-scoped, human-readable Markdown—no vector database and no hidden consolidation pipeline.
Release: https://github.com/chncaesar/opencode-plugin-memory/releases/tag/v0.2.0
Repository: https://github.com/chncaesar/opencode-plugin-memory
Top comments (0)