LLVM Backend
Expert skill for LLVM integration including IR generation, optimization passes, and native code emission
Best use case
LLVM Backend is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Expert skill for LLVM integration including IR generation, optimization passes, and native code emission
Teams using LLVM Backend should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/llvm-backend/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How LLVM Backend Compares
| Feature / Agent | LLVM Backend | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Expert skill for LLVM integration including IR generation, optimization passes, and native code emission
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# LLVM Backend Skill
## Overview
Expert skill for LLVM integration including IR generation, optimization passes, and native code emission.
## Capabilities
- Generate LLVM IR from high-level AST/IR
- Configure and run LLVM optimization passes
- Implement custom LLVM passes
- Handle LLVM type system mapping
- Generate debug information (DWARF)
- Configure target machine and code generation options
- Implement LLVM JIT (ORC, MCJIT) integration
- Handle cross-compilation target triples
## Target Processes
- code-generation-llvm.js
- jit-compiler-development.js
- debugger-adapter-development.js
- ir-design.js
## Dependencies
- LLVM C++ API
- llvm-sys bindings
- Inkwell (Rust LLVM bindings)
## Usage Guidelines
1. **Type Mapping**: Establish clear mapping between source types and LLVM types
2. **SSA Form**: Leverage LLVM's SSA form; generate clean IR and let LLVM optimize
3. **Debug Info**: Generate debug info from the start using DIBuilder
4. **Optimization Levels**: Test with -O0 first, then enable optimizations incrementally
5. **Target Configuration**: Abstract target-specific code behind target triple configuration
## Output Schema
```json
{
"type": "object",
"properties": {
"llvmVersion": { "type": "string" },
"targetTriple": { "type": "string" },
"optimizationLevel": {
"type": "string",
"enum": ["O0", "O1", "O2", "O3", "Os", "Oz"]
},
"passes": {
"type": "array",
"items": { "type": "string" }
},
"generatedFiles": {
"type": "array",
"items": { "type": "string" }
}
}
}
```Related Skills
backend-selector
Multi-backend comparison and selection skill for optimal hardware choice
redis-memory-backend
Redis backend for conversation state persistence and caching
process-builder
Scaffold new babysitter process definitions following SDK patterns, proper structure, and best practices. Guides the 3-phase workflow from research to implementation.
babysitter
Orchestrate via @babysitter. Use this skill when asked to babysit a run, orchestrate a process or whenever it is called explicitly. (babysit, babysitter, orchestrate, orchestrate a run, workflow, etc.)
yolo
Run Babysitter autonomously with minimal manual interruption.
user-install
Install the user-level Babysitter Codex setup.
team-install
Install the team-pinned Babysitter Codex workspace setup.
retrospect
Summarize or retrospect on a completed Babysitter run.
resume
Resume an existing Babysitter run from Codex.
project-install
Install the Babysitter Codex workspace integration into the current project.
plan
Plan a Babysitter workflow without executing the run.
observe
Observe, inspect, or monitor a Babysitter run.