Unifying AI Coding Assistant Configs with Ruler

Every AI coding tool wants its configuration in a different fucking place. Claude Code uses CLAUDE.md, OpenCode wants AGENTS.md, Cursor needs .cursor/rules/, Copilot expects .github/copilot-instructions.md. If you're using multiple tools, you're copy-pasting the same instructions everywhere like an idiot.

The Problem: Configuration Chaos

Look at this shit:

• Claude Code: ~/CLAUDE.md
• OpenCode: ~/AGENTS.md
• Cursor: .cursor/rules/ruler_cursor_instructions.md
• Copilot: .github/copilot-instructions.md
• Windsurf: .windsurf/rules/ruler_windsurf_instructions.md
• Aider: Both AGENTS.md AND .aider.conf.yml
• 20+ more tools, each with their own bullshit location

When I update my instructions, I have to remember to update ALL of them. Fuck that.

The Solution: Ruler Does the Distribution

Ruler maintains ONE source of truth in .ruler/ and automatically distributes to all tools. It's already built by Eleanor Berger (intellectronica), works perfectly, don't reinvent it.

# Install Ruler (it's npm now, not pip)
npm install -g @intellectronica/ruler

# Initialize in your home directory
cd ~
ruler init

# This creates:
# .ruler/
#   ├── AGENTS.md       (your main instructions)
#   └── ruler.toml      (config for which tools to update)

Critical Safety: Preventing Context Explosion

Here's how I fucked up and you can avoid it:

The safety protocol:

# GOLDEN RULE: Only these files belong in .ruler/
~/.ruler/AGENTS.md        # Primary source
~/.ruler/agents/*.md      # Agent modules  
~/.ruler/ruler.toml       # Configuration

# NEVER create these in .ruler/:
.windsurf/                # Tool outputs
.cursor/                  # Tool configs
*.backup                  # Temp files

# If contamination happens:
cd ~/.ruler
rm -rf .[a-z]*           # Remove hidden dirs
ruler revert             # Undo everything
ruler apply              # Re-apply clean

MCP Server Configuration

Ruler also manages Model Context Protocol servers. Define them once in TOML:

[mcp_servers.filesystem]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/home/wv3"]

[mcp_servers.git]  
command = "npx"
args = ["-y", "@modelcontextprotocol/server-git", "--repository", "."]

These get distributed to compatible tools automatically. No more maintaining separate MCP configs.

Results

Why This Matters

Mad aweomse frameworks are dropping like 2x a day, lots good, lots whateverr, some real frickin gems.

I want to stay on top of them and be able to quickly test ideas but using only claude code i realized I was limiting myself seriously. OpenCode is fucking sweet, that guy Dax knows how to build a TUI

Now I edit .ruler/AGENTS.md, run ruler apply, and all tools get updated. When a new AI tool drops next week, I add 3 lines to ruler.toml and it's integrated.

Implementation Notes

Shit that will bite you:

• File discovery is recursive - ALL .md files get included
• Order matters - files concatenate alphabetically
• Ruler NEVER writes outside your project root (except global config)
• Agent formats differ between tools - conversion scripts needed if you want to carry them over.
  • i just did opencode and claudecode so far

Don't create your own distribution system. Ruler fucking works.

Next is plugging all this back into my inter-agent messaging system i had built last week to make all my different claude instances talk to each other via tmux commands. excited to see how this looks leveraging all the smaller, cheaper (literally openrouter just has free models you use, they arent the most powerful but realizing that admin tasks around the homelab dont need giant coding models to get the job done. opencode provides solid prompting enough that these ffree models literally work good.)

also fwiw the free Grok Code Fast model on OpenRouter is snappy rn - get it while its hot. it feels like gpt-5 if it laid down the actual lines of code a bit heavier. really nnice time with it yesterday.