dream-skill
Memory consolidation skill for Claude Code that auto-triggers every 24 hours to keep AI agent memory clean and
About dream-skill
Dream-Skill is a GitHub-hosted skill that implements a four-phase memory consolidation process for Claude Code agents, inspired by Anthropic’s unreleased 'auto-dream' feature. It runs automatically once per day to prune outdated or contradictory memories, helping maintain coherence in long-running agent workflows.
The skill operates without requiring Anthropic feature flags or waiting for official rollout — it integrates directly into existing Claude Code agent setups. Its design mimics biological memory consolidation during sleep, applying periodic cleanup to reduce cognitive drift over time.
It is distributed as source code with no hosted service or UI; users must self-deploy and configure it within their Claude Code agent environment.
Key Features
- Replicates Anthropic's unreleased auto-dream functionality for Claude Code
- Four-phase memory consolidation process
- Auto-triggers every 24 hours without manual intervention
- Designed to reduce memory contradictions and improve long-term agent consistency
- Self-hosted and integrated directly into Claude Code agent workflows
Pros & Cons
Pros
- • Enables advanced memory hygiene before official Anthropic support arrives
- • No dependency on Anthropic feature flags or platform updates
- • Lightweight integration for teams already using Claude Code
Cons
- • Requires manual setup and configuration — no installer or guided onboarding
- • No documentation on compatibility with specific Claude Code versions or runtime environments
- • Lack of license information makes redistribution or enterprise use uncertain