Scan memory files, archive stale items, and merge duplicates into MEMORY.md.
npx skills add https://github.com/notque/vexjoy-agent --skill auto-dream
Please help me install this Agent Skill. Skill: auto-dream Source: https://github.com/notque/vexjoy-agent/tree/main/skills/meta/auto-dream Command: npx skills add https://github.com/notque/vexjoy-agent --skill auto-dream
This Skill prevents knowledge bloat and stale guidance by automatically scanning, deduplicating, and consolidating memory files into a clean, injection-ready payload for the next session.
Use Case: Run nightly maintenance to keep MEMORY.md accurate and reduce contradictions, so the agent carries fresh operational knowledge forward without accumulating outdated or redundant instructions.
Run the nightly dream job (dry-run first) using: ./scripts/auto-dream-cron.sh
High-intent search queries and answers about installing and using this skill.
Memory consolidation and deduplication is handled by scanning all memory files and recent session evidence to archive stale items and merge duplicates into an injection-ready payload. This prevents knowledge bloat and ensures fresh operational guidance is carried forward for the next session.
Safe memory cleanup enforces strict safety gates including never deleting files, applying atomic updates via tmp rename, and defaulting to dry-run mode. It flags conflicts for human review and caps maximum changes per cycle to prevent silent behavioral shifts.
Learning graduation promotes high-confidence, repeatedly observed entries from the sqlite learning db into permanent agent or skill files. This synthesis is capped and routed through a review branch rather than silently altering agent behavior.
Cron automation supports nightly background maintenance by scanning for stale or duplicate entries and curating an injection-ready payload. You can trigger the nightly dream job on-demand or via a cron script, starting with a dry-run to preview changes.
Memory consolidation limits include capping the number of changes or graduations per cycle to keep outputs focused. When conflicting entries are detected during scanning, the system flags them for human review rather than automatically resolving the discrepancy.
Chachamaru127Record session logs, decisions, and patterns to .claude/memory/ files.
ag-gridStore branch context or project learnings as persistent agent memory.
OmerZeyveliRecords validated UI R&D findings from complex debugging sessions into searchable, structured summaries.
bahayonghangAnalyze staged changes and draft Conventional Commits with safety scans.
