AI Agent Resources
Build Smarter AI Agents
AI coding assistants are changing how we write software. But most tools treat your codebase as just text files. They grep, they guess, they miss context. Your IDE knows better—it has indexed everything, understands every symbol, sees every relationship.
This page provides the tools to bridge that gap: MCP Steroid gives AI agents direct access to IntelliJ’s brain, and prompt skill files teach agents how to orchestrate complex tasks.
On this page: MCP Steroid · run-agent.sh · AI Agent Prompts · Connect
MCP Steroid
Give your AI agent a senior developer’s toolkit. MCP Steroid is an IntelliJ IDEA plugin that connects AI agents to the IDE’s runtime via the Model Context Protocol. Instead of reading text files and guessing at code structure, your AI assistant gets direct access to IntelliJ’s indexed code intelligence—the same power that makes IntelliJ the world’s most productive IDE.
The IDE has indexed everything. It knows the code better than any file search. Your agent can navigate code semantically with go-to-definition and find-usages. It can run refactorings that update imports and rename symbols project-wide. It sees the same errors and warnings IntelliJ shows. It queries the project model for modules, dependencies, and source roots. It can even debug applications—setting breakpoints, controlling sessions, and inspecting threads.
Agents submit Kotlin scripts via steroid_execute_code, and a review gate allows human approval
before execution. Vision tools capture screenshots with component trees and OCR, enabling agents
to click what they see.
Works with Any AI—Including Local Models
MCP Steroid isn’t locked to any AI vendor. Connect Claude Code, Codex CLI, Gemini CLI, or any MCP-compatible agent. Run local models via Ollama for complete privacy—your code never leaves your machine. No credits, no API costs, no vendor lock-in. As AI API prices rise, LocalAI becomes the smart hedge.
Quick Start:
- Install the MCP Steroid plugin in IntelliJ IDEA 2025.3+
- Open any project—the MCP server starts automatically
- Connect your AI agent using the URL from
.idea/mcp-steroid.md
# Claude Code
claude mcp add --transport http intellij-steroid "$(cat .idea/mcp-steroid.md)"
# Codex CLI
codex mcp add intellij-steroid "$(cat .idea/mcp-steroid.md)"
# Gemini CLI
gemini mcp add intellij-steroid "$(cat .idea/mcp-steroid.md)" --transport http --scope user
Homepage · GitHub · More Videos Tags: mcp, intellij, ai-coding
run-agent.sh
Orchestrate dozens of AI agents from your terminal. run-agent.sh is a shell-based framework that launches Claude, Codex, or Gemini agents in parallel – research, implementation, review, testing, and debugging agents all working on the same codebase simultaneously.
The entire framework is two files: run-agent.sh (a unified runner with full isolation and traceability) and THE_PROMPT_v5.md (a 13-stage orchestration workflow with 7 agent roles). Agents coordinate through a file-based append-only message bus – no databases, no servers. Every run is captured: prompts, stdout/stderr, PIDs, exit codes.
The orchestration lives outside your project sources, keeping your codebase clean. Combine with MCP Steroid for IDE-level quality gates – real inspections and refactorings before commit.
Quick Start:
# Download the runner
curl -sO https://run-agent.jonnyzzz.com/run-agent.sh
chmod +x run-agent.sh
# Download the orchestration prompt
curl -sO https://run-agent.jonnyzzz.com/THE_PROMPT_v5.md
Homepage · GitHub · Blog Post Tags: ai-agents, ai-coding, orchestration, dev-tools
AI Agent Prompts
Prompt skill files teach AI agents how to orchestrate complex tasks. Copy them to your project root to enable multi-agent coordination, task decomposition, and CLI tool integration. These follow the llms.txt specification for AI-readable content.
RLM.md — Recursive Language Models. Actionable instructions for breaking large tasks into manageable pieces and handling contexts that exceed token limits. The core pattern for making AI agents work on real codebases.
MULTI-AGENT.md — Agent orchestration patterns. How to coordinate multiple sub-agents for complex tasks. Includes templates for parallel execution, cross-validation, and sequential pipelines.
GIT-FORK.md — Git Fork Pattern. Create independent git checkouts with shared objects. A lightweight alternative to git worktree that lets multiple agents work on the same repo simultaneously.
CLAUDE-CODE.md — Using Claude Code CLI as a sub-agent. Full tool access, web search, parallel execution. The most capable agent for complex coding tasks.
CODEX.md — Using Codex CLI as a sub-agent. Ideal for image analysis, focused code review, and parallel workloads. Supports PDF and image input.
GEMINI.md — Using Gemini CLI as a sub-agent. Provides alternative perspective for cross-validation and excels at long-context tasks.
Download all:
curl -sO https://jonnyzzz.com/RLM.md
curl -sO https://jonnyzzz.com/MULTI-AGENT.md
curl -sO https://jonnyzzz.com/GIT-FORK.md
curl -sO https://jonnyzzz.com/CLAUDE-CODE.md
curl -sO https://jonnyzzz.com/CODEX.md
curl -sO https://jonnyzzz.com/GEMINI.md
Connect
Building with AI agents? Follow me for updates on MCP tools, multi-agent patterns, and developer tooling: