# Memory Search

> Search Ruflo memory with semantic retrieval, keyword/semantic union, or optional SmartRetrieval reranking

- Skill: `ruvnet/memory-search-2` (Agent Skill)
- Install (CLI): `npx get-skill add ruvnet/memory-search-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ruvnet/memory-search-2/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: ruvnet (https://skillmd.com/u/ruvnet)
- Updated: 2026-11-11
- Page: https://skillmd.com/skills/ruvnet/memory-search-2

---


# Memory Search

Search Ruflo's stored memories. Select a strategy supported by the Ruflo CLI or MCP tool, and report which one actually ran.

## Strategy selection

- **Default:** semantic retrieval through `memory search` or MCP `memory_search`.
- **`--hybrid`:** CLI keyword + semantic result union, deduplicated and sorted by score. This is not the `ruvector search` CLI and does not claim Graph RAG or sparse/dense RRF.
- **`--smart`:** request query expansion, RRF across variants, recency weighting, and MMR diversity. The MCP tool can fall back to plain search if the installed memory package lacks `smartSearch`; inspect the returned fallback note before claiming SmartRetrieval ran.
- **Cross-namespace:** MCP `memory_search_unified` merges and deduplicates results from multiple namespaces.

If the user requests `--graph-rag`, explain that this plugin has no supported text-query Graph RAG command. Offer `--smart` semantic retrieval as a clearly labeled fallback; do not claim graph traversal.

## Steps

1. Parse the query, optional namespace, result limit, and strategy flags. Omit `--namespace` to search all namespaces.
   For CLI commands, treat the query as untrusted shell input: use POSIX single-quote escaping for embedded apostrophes and reject embedded newlines. Do not interpolate it inside double quotes.
2. Run one of the supported paths:

   **Semantic (default)**
   ```bash
   npx @claude-flow/cli@latest memory search --query 'QUERY' --limit 10
   ```
   MCP: `mcp__plugin_ruflo-core_ruflo__memory_search({ query: "QUERY", limit: 10 })`

   **Keyword + semantic union (`--hybrid`)**
   ```bash
   npx @claude-flow/cli@latest memory search --query 'QUERY' --type hybrid --limit 10
   ```
   Add `--namespace NAMESPACE` only when the user requests one.

   **Smart retrieval (`--smart`)**
   ```bash
   npx @claude-flow/cli@latest memory search --query 'QUERY' --smart --limit 10
   ```
   MCP: `mcp__plugin_ruflo-core_ruflo__memory_search({ query: "QUERY", smart: true, limit: 10 })`

   **Cross-namespace bridge search**
   `mcp__plugin_ruflo-core_ruflo__memory_search_unified({ query: "QUERY", limit: 10 })`

3. Only when the question is about how earlier tasks like this one went, also ask for their validated episodes:
   `mcp__plugin_ruflo-core_ruflo__agentdb_context-synthesize({ query: "QUERY", maxEntries: 10 })`
   It is not a document synthesizer: it summarizes validated task episodes (task, reward, success) and answers `no-memories` or `no-eligible-episodes` for ordinary records, which is correct, not missing data.
4. Present the returned entries with namespace attribution. Distinguish raw `similarity` from SmartRetrieval's `rankingScore`; neither is a probability of truth. State when SmartRetrieval fell back to plain search.

## Namespace guide

| Namespace | Typical content |
|-----------|-----------------|
| `patterns` | Reusable approaches |
| `tasks` | Task context |
| `solutions` | Resolved problems |
| `feedback` | User corrections |
| `security` | Security notes |
| (omit) | All available namespaces |

