What is the primary purpose of the Retrieval Policy skill?

It acts as a decision-making engine that determines the best way to search an agent's memory based on the query's intent and available search capabilities.

Can I see why a specific search method was chosen?

Yes, every retrieval operation returns an explainability payload detailing the layers tried, fallbacks triggered, execution time, and the final confidence score.

How does it handle failed search attempts?

It uses predefined fallback chains to move from high-complexity layers, like Topics or Hybrid search, to simpler ones, like Agentic TOC navigation, until a result is found.

What are the different query intents it classifies?

It classifies queries into four categories: Explore (broad discovery), Answer (fact-finding), Locate (exact matches), and Time-boxed (date-specific constraints).

Does this skill perform the actual search operation?

No, it is a policy and routing engine. It should be used in conjunction with specific search skills like vector-search or bm25-search to execute the actual data retrieval.

Intelligent Memory Retrieval Policy

Name: Intelligent Memory Retrieval Policy
Author: SpillwaveSolutions

bySpillwaveSolutions

•

Data Science & ML

Optimizes AI agent memory searches by classifying query intent and routing requests through the most efficient retrieval layers.

The Retrieval Policy skill serves as the central decision-making engine for agent memory systems, functioning as a 'brainstem' that determines the optimal path for information retrieval. It analyzes incoming queries to detect intent—categorizing them as exploratory, factual, location-based, or time-sensitive—and routes them through various capability tiers including BM25, vector embeddings, and agentic TOC navigation. By implementing automated fallback chains and providing detailed explainability payloads, it ensures that Claude can reliably find the most relevant information while maintaining full transparency over why specific search methods were chosen.

Key Features

01Automated query intent classification for optimized search strategies

024 GitHub stars

03Dynamic tier-based routing across BM25, Vector, and Agentic layers

04Configurable stop conditions including max depth, nodes, and timeouts

05Predefined fallback chains to handle layer failures or zero-result scenarios

06Detailed explainability payloads reporting layers tried, time taken, and confidence

Use Cases

01Determining the most efficient search layer for complex natural language queries

02Implementing robust memory retrieval with automatic fallbacks for high-reliability agents

03Debugging search performance using detailed execution and intent classification reports

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add spillwavesolutions/agent-memory retrieval-policy

For use in Claude.ai and ChatGPT

Download Skill