Neural Train
- 653 installs
- 67k repo stars
- Updated August 4, 2026
- ruvnet/ruflo
neural-train is an agent skill that trains SONA and MicroLoRA neural patterns from successful task completions for developers who want agent workflows to learn coordination, edit, and task patterns.
About
neural-train is a ruvnet/ruflo skill that trains SONA plus MicroLoRA neural patterns from successful task completions, executing the DISTILL and CONSOLIDATE phases of a documented four-step pipeline. CLI flags include --pattern-type coordination|edit|task, --epochs N, and --microlora, routed through Claude-flow MCP tools such as neural_train, neural_status, neural_patterns, neural_predict, neural_optimize, neural_compress, and intelligence trajectory hooks. Developers invoke neural-train inside terminal or agent workflows when prior runs produced high-quality edits or coordination traces worth compressing into reusable patterns. The skill targets small learned behaviors for coding agents rather than full foundation-model fine-tuning. Reach for neural-train after collecting successful trajectories you want agents to replay with lower variance on similar tasks. Intelligence trajectory hooks cover pretrain, trajectory-start, trajectory-step, and trajectory-end events so multi-step agent runs feed the DISTILL phase with coordination, edit, or task pattern labels. neural_status and neural_patterns MCP calls report training progress before predict and optimize reuse consolidated weights.
- Fine-tunes LLMs on custom datasets using neural network training loops
- Runs locally or in cloud with one-command training sessions
- Supports multiple model architectures and dataset formats
- Integrates with agent coding tools for automated model improvement
- Tracks metrics including loss curves, accuracy, and epoch progress
Neural Train by the numbers
- 653 all-time installs (skills.sh)
- +10 installs in the week ending Jul 26, 2026 (Skillselion tracking)
- Ranked #1,478 of 16,546 AI & Agent Building skills by installs in the Skillselion catalog
- Data as of Aug 5, 2026 (Skillselion catalog sync)
npx skills add https://github.com/ruvnet/ruflo --skill neural-trainAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Installs | 653 |
|---|---|
| repo stars | ★ 67k |
| Last updated | August 4, 2026 |
| Repository | ruvnet/ruflo ↗ |
How do you train agent neural patterns from tasks?
Fine-tune small language models directly from the terminal or within agent workflows.
Who is it for?
Developers using Claude-flow who want to distill successful agent trajectories into SONA or MicroLoRA coordination and edit patterns.
Skip if: Developers fine-tuning large foundation models on generic corpora outside Claude-flow neural MCP tooling and trajectory hooks.
When should I use this skill?
User asks to neural-train, distill agent patterns, run MicroLoRA, or consolidate successful task trajectories into learned behaviors.
What you get
Trained SONA/MicroLoRA pattern weights, neural_status report, and consolidated patterns for predict and optimize MCP calls.
- Trained neural patterns
- neural_status report
- Consolidated pattern store entries
By the numbers
- Runs DISTILL and CONSOLIDATE phases of a documented 4-step pipeline
- Supports 3 pattern-type modes: coordination, edit, and task
Files
Neural Training
Train and consolidate neural patterns. Implements the DISTILL and CONSOLIDATE phases of the 4-step intelligence pipeline.
When to use
- After completing a successful task — capture what worked.
- After accumulating ≥10 task completions — run consolidation to fold patterns into long-term storage.
- When training a new domain — create a MicroLoRA adapter for it.
Standard flow (DISTILL)
1. Check current neural status — mcp__claude-flow__neural_status. 2. Start a trajectory — mcp__claude-flow__hooks_intelligence_trajectory-start with the task context. 3. Record steps — for each significant action, mcp__claude-flow__hooks_intelligence_trajectory-step. 4. End trajectory — mcp__claude-flow__hooks_intelligence_trajectory-end with verdict: pass|fail|partial. 5. Learn from the trajectory — mcp__claude-flow__hooks_intelligence_learn. 6. Train patterns — mcp__claude-flow__neural_train with --pattern-type coordination --epochs 10. 7. Store patterns — mcp__claude-flow__hooks_intelligence_pattern-store. 8. Verify — mcp__claude-flow__neural_patterns to confirm.
SONA adaptation (single-domain, <0.05ms)
For real-time micro-adaptation:
mcp tool call ruvllm_sona_create --json -- '{"domain": "coding"}'
mcp tool call ruvllm_sona_adapt --json -- '{"feedback": {"score": 0.9, "trajectory": "..."}}'MicroLoRA adaptation (multi-domain)
When you have ≥3 distinct domains, create a MicroLoRA adapter per domain rather than overloading SONA:
# Create the adapter
mcp tool call ruvllm_microlora_create --json -- '{"domain": "frontend"}'
# Adapt with feedback
mcp tool call ruvllm_microlora_adapt --json -- '{"adapter": "frontend", "feedback": {...}}'
# CONSOLIDATE phase: apply EWC++ on weight deltas to prevent catastrophic forgetting
mcp tool call ruvllm_microlora_adapt --json -- '{"adapter": "frontend", "consolidate": true}'The --consolidate flag is the EWC++ trigger. Without it, fresh training overwrites older domains.
CONSOLIDATE phase (separate from training)
After every ~10 trajectory completions, run a full consolidation pass:
mcp tool call agentdb_consolidate --json
mcp tool call neural_compress --json # storage efficiencyThis folds patterns into long-term storage under EWC++ semantics.
Bootstrapping from scratch
If the system has no learned patterns yet:
mcp tool call hooks_pretrain --json -- '{"modelType": "moe", "epochs": 10}'
mcp tool call hooks_build-agents --json -- '{"agentTypes": "coder,tester"}'hooks_pretrain writes to the patterns (plural) namespace — distinct from the pattern (singular) ReasoningBank target. See ruflo-agentdb ADR-0001 for the namespace convention.
Reset (testing only)
To wipe intelligence state (e.g., for benchmarking):
mcp tool call hooks_intelligence-reset --jsonCLI alternatives
npx @claude-flow/cli@latest neural train --pattern-type coordination --epochs 10
npx @claude-flow/cli@latest neural patterns --list
npx @claude-flow/cli@latest neural status
npx @claude-flow/cli@latest neural compress
npx @claude-flow/cli@latest hooks pretrain --model-type moe --epochs 10
npx @claude-flow/cli@latest hooks build-agents --agent-types coder,testerRelated skills
Setup Matt Pocock SkillsScaffold the per-repo configuration that Matt Pocock’s engineering agent skills rely on so they understand the issue tracker, triage labels, and domain documentation la495k196k
Lark AppsConnect Claude, Cursor or custom agents directly to Lark (Feishu) for messaging, document automation, approval workflows and enterprise data access.402k
CavemanSlash token usage by roughly 75% while keeping every technical detail intact when working with Claude Code, Cursor or similar agents.391k94.7k
Lark Skill MakerQuickly turn any Lark/Feishu OpenAPI call or multi-step workflow into a reusable agent skill with its own SKILL.md.386k16k
Running Claude Code Via Litellm CopilotRun Claude Code at a fraction of the cost by routing requests through LiteLLM to the GitHub Copilot Chat API.270k72
Codex PetGenerate a complete Codex Pet spritesheet and metadata from one reference image without needing an OpenAI key or Codex Pro.246k10How it compares
Pick neural-train for Claude-flow trajectory distillation into SONA/MicroLoRA patterns instead of manual prompt tuning without learned weights.
FAQ
What pipeline phases does neural-train run?
neural-train runs the DISTILL and CONSOLIDATE phases of a four-step pipeline, training SONA and MicroLoRA neural patterns from successful task completions captured in agent workflows.
Which CLI flags does neural-train accept?
neural-train accepts --pattern-type coordination|edit|task, --epochs N, and --microlora as argument-hint flags routed to mcp__claude-flow__neural_train and related neural MCP endpoints.
Which MCP tools power neural-train?
neural-train uses Claude-flow neural MCP tools including neural_train, neural_status, neural_patterns, neural_predict, neural_optize, neural_compress, plus intelligence trajectory hook endpoints for multi-step runs.