doctor

Inspect and prune stale nodes in the memory graph after explicit confirmation.

20|2|Updated May 16, 2026
One-click install
npx skills add https://github.com/reddb-io/red-skills --skill doctor-reddb-io
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: doctor
Source: https://github.com/reddb-io/red-skills/tree/main/plugins/memory/skills/core/doctor
Command: npx skills add https://github.com/reddb-io/red-skills --skill doctor-reddb-io

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Over time, an agent memory graph accumulates nodes that are never recalled and have not been accessed in months, cluttering recall results. This Skill identifies those stale nodes and removes them safely, but only after the user explicitly confirms the deletion. ## Core Features & Use Cases - Stale-Node Scan: Lists nodes last accessed beyond a configurable threshold (90 days by default) that have never been recalled, without deleting anything. - Confirmed Pruning: Re-lists candidates and requires a typed confirmation before deleting; pinned nodes (importance >= 0.8) are never flagged as stale. - Use Case: After months of agent operation, run a memory health check to see which old decisions or facts nobody ever reads, review the list, and prune them to keep the graph clean. ## Quick Start Ask the agent to run a memory doctor health check and show which memory nodes are stale.

Frequently Asked Questions about doctor

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I clean up stale memory in an agent memory graph?▼

Run the doctor command to list stale nodes, meaning nodes not accessed within the threshold and never recalled. Review the list, then re-run with the prune flag and confirm to delete them.

What counts as a stale node in the memory graph?▼

A node is stale when its last access is older than the threshold, 90 days by default, and its access count is zero, meaning it was never recalled. Nodes pinned with importance of 0.8 or higher are never considered stale.

Can memory pruning run automatically without confirmation?▼

No. Pruning is destructive and always requires explicit confirmation. In non-interactive shells the prune flag refuses to run unless a yes flag is passed, which should only happen after the user has confirmed.

Does the memory doctor work in markdown-only mode?▼

No. The doctor requires graph mode because it inspects the memory graph and its access overlay. If memory is uninitialized or markdown-only, the Skill reports that and stops.

Is the memory doctor available through MCP?▼

Only the read-only half is exposed over MCP as memory_doctor with a stale_days argument. Pruning remains a confirmed CLI operation and is intentionally not exposed over MCP.