intelligence-transfer

name	intelligence-transfer
description	Transfer learned patterns from another project on disk into this one -- the project-to-project pattern transfer that hooks_transfer performs
argument-hint	--source <project-path> [--filter <type>] [--min-confidence <0-1>]
allowed-tools	mcp__ruflo__hooks_transfer mcp__ruflo__hooks_intelligence_pattern-search mcp__ruflo__hooks_intelligence_pattern-store mcp__ruflo__neural_patterns mcp__ruflo__neural_status Bash

Intelligence Transfer

Project-to-project pattern sharing. Reads the learned patterns from another project's memory store on disk and reports what would transfer into this one.

What hooks_transfer actually does

hooks_transfer is a local, on-disk operation — it reads a source project's memory store (<sourcePath>/.claude-flow/memory/store.json) and counts the patterns it can transfer, grouped by type (file-patterns, task-routing, command-risk, agent-success). There is no IPFS, no Pinata, no CID, and no network call — this skill previously documented a surface the tool never had (ADR-0293 D10).

Parameters

• sourcePath (required) — filesystem path to the source project.
• filter (optional) — only transfer pattern types whose key contains this string (e.g. routing).
• minConfidence (optional, default 0.7) — minimum confidence threshold.

Workflow

# Inspect what's stored locally first
mcp tool call neural_patterns --json -- '{"action": "list"}'

# Transfer patterns from a sibling project on disk
mcp tool call hooks_transfer --json -- '{"sourcePath": "/path/to/peer-project"}'

# Filter to a single pattern family, with a confidence floor
mcp tool call hooks_transfer --json -- '{"sourcePath": "/path/to/peer-project", "filter": "routing", "minConfidence": 0.8}'

On success the response is:

{ "success": true, "sourcePath": "...", "transferred": { "total": 2, "byType": { "task-routing": 1, "agent-success": 1, "...": 0 } }, "dataSource": "source-project" }

If the source has no patterns (empty or nonexistent), the response is honest — it does NOT fabricate demo data (ADR-0293 D2):

{ "success": false, "message": "No patterns found in source project", "sourcePath": "...", "transferred": 0 }

When to use this skill

• Before a fresh project starts — pull the relevant patterns from a parent project on the same machine so the new project's agents start with prior knowledge instead of cold.
• Consolidating a monorepo or fleet — fold a sibling project's learned patterns into a shared project.

When NOT to use

• Across machines / over a network — hooks_transfer is local-disk only. There is no remote/IPFS transport in this build. Copy the source project's .claude-flow/memory/store.json to the target machine first, then run this against the local copy.
• For sensitive patterns — strip PII (use aidefence_has_pii first) before sharing a store between projects.

Caveats

• This reads only the intelligence-side memory store; it does NOT export AgentDB rows. To ship full memory, use the agentdb_* / memory_export tools (out of scope here).
• The counts reflect the source project's actual entries — an empty source yields success: false, never a fabricated count.

Related

• ruflo-agentdb ADR-0001 §"Namespace convention" — the pattern namespace
• neural-train skill — produces the patterns this skill transfers