Context Keeper

Solves the problem of Claude forgetting design decisions and constraints across conversations by storing structured project memory as JSON files in your repo. You get ten tools to record decisions with required rationale fields (problem, why_chosen, tradeoffs), pipeline workflows with ordering, and constraints with enforcement levels. The get_context tool pulls related entries via ID links, and hooks inject summaries at session start plus snapshot context before Claude Code compaction to catch anything lost. Enforces minimum character counts on reasoning fields so agents can't write lazy one-liners. All data lives in .context/ as human-editable files. Useful when working on projects that span multiple conversations where Claude needs to remember why you chose approach X over Y.