idea2code

Turn a research idea into running code. One command, one experiment.

"Talk is cheap. Show me the code." But now in the era of AI coding, the reverse might be true — your ability to create a new idea is far more important than implementing it.

This tool bridges that gap. You bring the idea. It writes the code.

/idea-iter try attention gates in the decoder

What it does

You describe an idea. The agent:

• Finds relevant papers to inform the implementation
• Reads your codebase to understand the architecture
• Makes surgical code edits to implement your idea
• Commits, pushes, and launches the experiment
• Returns immediately so you can start the next idea

Install

Option A: npx (requires Node.js)

npx idea2code

Option B: git clone (no Node.js needed)

git clone https://github.com/haoyudong-97/idea2code.git /tmp/idea2code && \
  rm -rf ~/.claude/skills/idea-iter ~/.claude/skills/check-experiments ~/.claude/skills/combine-findings && \
  cp -r /tmp/idea2code/skill/idea-iter ~/.claude/skills/ && \
  cp -r /tmp/idea2code/skill/check-experiments ~/.claude/skills/ && \
  cp -r /tmp/idea2code/skill/combine-findings ~/.claude/skills/ && \
  for s in idea-iter check-experiments combine-findings; do \
    cp -r /tmp/idea2code/skill/research_agent ~/.claude/skills/$s/; \
  done && \
  rm -rf /tmp/idea2code && \
  echo "Done! Skills installed."

Requirements

• Claude Code installed
• Python 3.10+
• Git

Uninstall

rm -rf ~/.claude/skills/idea-iter ~/.claude/skills/check-experiments ~/.claude/skills/combine-findings

Usage

cd your-project && claude

/idea-iter try attention gates in the decoder       # idea -> papers -> code -> launch
/idea-iter --auto increase batch size to 4          # skip confirmation, launch directly
/check-experiments                                   # collect results when training finishes
/combine-findings https://arxiv.org/abs/2401...      # integrate a specific paper

Run multiple iterations in parallel — each gets its own git branch and checkpoint:

/idea-iter add attention gates to decoder        -> iter 1 launched
/idea-iter increase batch size to 4              -> iter 2 launched
/idea-iter try cosine annealing schedule         -> iter 3 launched

/check-experiments                               -> collects all finished results

How It Works

Your idea
    ↓
Find relevant papers (arXiv API + WebSearch, top 10 with full text)
    ↓
Read your codebase, understand the architecture
    ↓
Implement the idea (surgical code edits via Agent)
    ↓
Commit to git branch (iter/1-attention-gates)
    ↓
Launch experiment (GPU-aware, local or remote SSH)
    ↓
Return immediately — start next idea

State tracking happens automatically via hooks — every git commit, deploy launch, and experiment completion updates state.json and progress.md without manual commands.

License

MIT