memory-navigation

Routes memory retrieval across search_evidence, graph_resolve, and list_recent entry points.

2.9k|733|Updated Mar 12, 2026
One-click install
npx skills add https://github.com/zts212653/clowder-ai --skill memory-navigation
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: memory-navigation
Source: https://github.com/zts212653/clowder-ai/tree/main/cat-cafe-skills/memory-navigation
Command: npx skills add https://github.com/zts212653/clowder-ai --skill memory-navigation

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

When working with a persistent memory system, agents often default to one search tool and waste effort on low-hit queries. This Skill provides a decision tree for choosing the right memory entry point—semantic search, graph resolution, or recent-activity scanning—based on how much prior context you have.

Core Features & Use Cases

  • Three-entry routing decision tree: Choose between search_evidence (concept keywords), graph_resolve (precise anchors), and list_recent (zero-prior scanning) based on your starting context.
  • Noise control parameters: Guidance on graph depth limits, relation filters, time windows, and search modes to prevent fan-out explosion and irrelevant results.
  • Low-hit recovery: Recognizes the memory navigation nudge signal from failed searches and redirects to alternative entry points instead of repeated blind querying.
  • Use Case: After a context compaction event, you remember discussing a feature but have no exact anchor. The Skill directs you to list_recent with a 7-day window to rebuild context quickly.

Quick Start

Ask the agent to help you find a past discussion you vaguely remember by routing through the right memory entry point.

Frequently Asked Questions about memory-navigation

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

FAQPage Schema
How do I choose between search_evidence, graph_resolve, and list_recent?

Use graph_resolve when you have a precise anchor like a feature ID, search_evidence when you have concept keywords, and list_recent when you have no prior context and just want to scan recent activity. The decision tree in the Skill maps each starting condition to the right entry point.

What should I do when search_evidence returns low-hit results?

Treat the memory navigation nudge in the search payload as a signal to switch entry points. Try graph_resolve with a nearby anchor or list_recent with a time window instead of repeatedly rephrasing keyword queries.

How do I prevent graph_resolve from returning too many results?

Keep the depth parameter at 1 and never exceed 3, since deeper traversal causes fan-out explosion. Filter relations to subsets like wikilink, feature_ref, or related_to, and start with high-confidence anchors before expanding.

When should I not use memory navigation routing?

Skip routing when you already have a precise anchor and can read the source file directly, or when you need code symbol lookup, which belongs to Grep or LSP tools. Full-text matching is also better handled by Grep.

Can graph_resolve and list_recent access private memory collections?

No, their MCP schemas do not accept callerCollections or collections parameters and only read public and internal collections. Accessing private or restricted collections requires calling the HTTP API directly with server-side ACL.