Claude Auto Dream - Setup & Health Check

Claude Auto Dream adds incremental memory extraction and session-end consolidation to Claude Code through a separate open-source plugin. This companion skill helps your agent install it correctly, choose sensible settings, and work out why a memory did or did not appear.

The runtime plugin is freely available under MIT at https://github.com/MeCKodo/claude-auto-dream. You are paying for the setup and diagnostic workflow, not exclusive access to that code.

What you get

• Guided installation with checks for the active plugin version and enabled hooks.
• Provider configuration review without dumping your API key or environment.
• Independent tuning of per-turn extraction and session-end consolidation.
• Diagnosis of missing configuration, wrong project paths, cursor state, locks, skip conditions, and provider errors.
• Verification that distinguishes a local check, a mock run, a real provider call, and an actual hook-driven memory update.

How it works

Ask your agent to set up the plugin or explain a failure. The skill starts from the actual local installation and configuration, makes a focused change when authorized, and reports what the evidence proves. It uses isolated synthetic checks before touching real project memories where possible.

Requirements

• Claude Code as the target runtime, with plugin support.
• macOS or Linux, Bash, and a supported Node.js version.
• An agent with terminal and local filesystem access; a browser-only chat cannot perform installation.
• A separately configured LLM provider and a tool-capable model. API usage is billed by that provider.

Scope and limitations

The automatic hooks require Claude Code. This listing does not add automatic memory to Cursor, Copilot, or every agent advertised by the marketplace. Current guidance is based on upstream v1.1.1 at commit 9bfa94139ba9b3554c7bc191b3a3a9ae2150a035; the skill instructs the agent to inspect a different installed version before relying on version-specific behavior.

Direct Anthropic compatibility in the reviewed runtime needs further verification. OpenAI-compatible providers also require a real tool-call test for the selected endpoint and model. Local mock verification does not prove a paid provider or current Claude Code hooks work end to end.

The plugin can read local memories and transcripts and send relevant content to your chosen model provider during actual runs. Memory changes depend on the model and input. No perfect recall, fixed API cost, hosted service, or human support SLA is included.