Lock in your definition of done before a single line of code is written.
kickoff is a Claude Code plugin that runs a structured project intake, asks 7 high-impact questions, and generates a ready-to-go CLAUDE.md so Claude has full context from session one. Works for new projects and for documenting existing codebases.
Most AI coding sessions drift because the goal was never clearly defined. kickoff enforces a definition-of-done gate: Claude cannot proceed until you've answered what "done" looks like, concretely and verifiably. The output is a CLAUDE.md that travels with your project and makes every future Claude session faster and more accurate.
It also introduces an Open Decisions pattern: anything you're not sure about gets deferred into a checklist rather than silently guessed. Claude resurfaces those decisions at the start of every session until they're resolved.
- Asks 7 focused intake questions — definition of done first, always
- Defers unknowns into an Open Decisions checklist so ambiguity is tracked, not hidden
- Proposes a scoped Session 1 goal before writing any code
- For existing projects: scans the codebase first and pre-fills what it can; shows diffs before overwriting
- Generates a
CLAUDE.mdwith type-specific sections for: data analysis, ML, web app/dashboard, automation, MCP server, API/backend, CLI, GAS/Sheets - Optionally sets up Python
venvor Nodepackage.json - Optional: initializes OpenWolf memory if installed
- Optional: creates an Obsidian project index if the Obsidian MCP is available
# Step 1 — add the kickoff marketplace
/plugin marketplace add babiskon/kickoff
# Step 2 — install the kickoff plugin
/plugin install kickoff@kickoffJust describe what you're building — kickoff activates automatically:
"let's start a new project from scratch"
"I need to build a data pipeline"
"kick off a new Python script"
"spin up a Streamlit dashboard"
"add CLAUDE.md to this existing repo"
"document this project for Claude"
"set it up properly from the start"
You'll be walked through 7 questions, then Claude writes the CLAUDE.md and optionally sets up your environment.
You: let's start a new project from scratch
Claude: What does "done" look like for this project?
Be specific: what output exists, where does it go, and how do we know it works?
You: a Python script that reads a CSV, runs some analysis, and writes a summary to Google Sheets
Claude: Got it. Now a few more questions:
2. Project name?
3. Type: (a) data analysis...
...
[after answers]
Claude: Here's what I suggest we accomplish today: a working script that reads the
CSV and prints a summary to stdout — no Sheets integration yet.
Does that sound right?
You: yes
Claude: [creates ~/projects/my_analysis/, writes CLAUDE.md, sets up venv]
Done. CLAUDE.md written. Open Decisions: 0.
Suggest: open the folder in your IDE and point the interpreter at venv/.
- Claude Code (any version)
- No external services required for core functionality
- Python venv setup requires Python 3.x
- Node.js setup requires Node.js
- OpenWolf integration requires the
openwolfCLI (optional) - Obsidian integration requires the Obsidian MCP server (optional)
MIT — see LICENSE.