What problem does it solve? Accumulated session learnings pile up in a backlog and never reach long-term memory without a controlled review process. This Skill provides a human-in-the-loop workflow that triages, resolves, and applies proposed memory updates so nothing enters memory unapproved and nothing reviewed gets lost. ## Core Features & Use Cases - Automated triage: A rubric, a judge model, and an in-code safety floor partition proposals into apply, review, and drop verdicts, with kind: RULE items never applied automatically. - Evidence-based resolution: Duplicates are merged, contradictions resolved by source date, and checkable claims verified against code or transcripts before writing. - Auditable decision record: Every decision is marked processed in the proposal file, receipts and rejection logs are written, and memory changes are reversible via git revert. - Use Case: After a week of coding sessions, run the skill to review a 190-item dream proposal: uncontroversial facts apply automatically, duplicates merge, and only genuine policy questions reach you. ## Quick Start Ask the agent to review and apply the pending memory updates from the learnings backlog using the dream-remember workflow.
