init

Set up eyeroll for the current project.

Step 1: Check if eyeroll is installed

eyeroll --help

If not found, ask the user which backend they want:

Use AskUserQuestion with these options:

• Gemini (Recommended) — Google Gemini Flash API, best quality, requires API key or credentials.json
• OpenAI — GPT-4o, requires API key
• Ollama — Local models, private, no API key needed

Then install:

• Gemini: pip install eyeroll[gemini]
• OpenAI: pip install eyeroll[openai]
• Ollama: pip install eyeroll

Step 2: Run eyeroll init

eyeroll init

This is interactive — it asks for the backend choice and API key. Let the user interact with it directly.

Step 3: Generate codebase context

After setup, explore the codebase and write .eyeroll/context.md:

1. Read CLAUDE.md, README.md, pyproject.toml or package.json
2. Run git log --oneline -10 and git branch --show-current
3. Run ls on root and key source directories
4. Skim 2-3 key source files

Write .eyeroll/context.md with this structure — keep it under 80 lines:

# Project: <name>
## Stack
<language, framework, key libraries>
## Structure
<directory layout>
## Key Files
- <file>: <purpose>
## Recent Activity
<last 5-10 commits, current branch>
## Notes
<anything relevant — conventions, known issues>

Also write .eyeroll/context_meta.json with the generation timestamp:

{"generated_at": "<current UTC ISO timestamp>"}

This lets eyeroll detect when the context goes stale (newer commits than the context).

Note: If the project already has CLAUDE.md, AGENTS.md, or similar coding tool context files, eyeroll auto-discovers those at watch-time — .eyeroll/context.md is only needed as a fallback when no such files exist.

Step 4: Confirm

Tell the user:

• eyeroll is ready
• .eyeroll/context.md has been generated (with staleness tracking)
• Codebase context is auto-discovered from CLAUDE.md, AGENTS.md, etc. if present
• They can run /eyeroll:watch <video> to analyze a video