Autopilot, step by step, or /feature?
Three ways to use nanostack. Pick the one that matches how you work.
The four workflows
| Workflow | Command | What happens | You decide |
|---|---|---|---|
| Autopilot | /think --autopilot | Full sprint runs end to end. Agent moves through every phase automatically. | Scope (before it starts) and review flags (when the agent is not confident). |
| Step by step | /think, then /nano, then /review, etc. | You trigger each phase manually. Agent waits for your go-ahead between phases. | Everything. You control the pace and can skip or repeat phases. |
| Feature | /feature | Skips /think diagnostic. Goes straight to planning for an existing project. | The feature description. Agent handles the rest. |
| Manual | Any skill directly | Run a single skill in isolation. No pipeline, no artifacts passed between phases. | Everything. Useful for one-off tasks. |
When to use each
Autopilot: first time building something new
Best for greenfield work where you want the full pipeline: scope challenge, plan, build, review, security, QA, ship. You describe the outcome,/think --autopilot runs the entire sprint, and you intervene only when the agent flags something it cannot decide on its own.
> Build a CLI tool that converts CSV files to JSON. Support stdin piping. > /think --autopilot
Good for: new projects, prototypes, features you can fully describe upfront.
Step by step: when you want control
Run each phase manually. This is useful when you want to review the plan before any code is written, or when you want to iterate on a specific phase without re-running the entire sprint.
> /think (review the scope challenge, adjust) > /nano (review the plan, tweak file list) > /review (read the review, fix one thing manually) > /ship
Good for: complex features, unfamiliar codebases, learning how each phase works.
Feature: adding to an existing project
When the project already exists and you want to add something specific,/feature is the fastest path. It reads your existing config and artifacts, skips the /think diagnostic (since the project is already scoped), and goes straight to planning.
/feature Add email notifications when a habit streak reaches 7 days
Good for: adding features, fixing bugs with known scope, iterating on an existing codebase.
Manual: quick fixes and one-off tasks
Sometimes you just need a code review, or a security check, or to ship what you already wrote. Run any skill directly without going through the pipeline.
> /review # review current changes > /security # audit before merging > /ship # commit and open PR
Good for: code you wrote yourself, quick audits, shipping work from a previous session.
Switching mid-sprint
You can start in autopilot and switch to step by step at any point. If the agent is running and you want to take over, just interrupt and run the next phase manually. Artifacts from completed phases are already saved, so nothing is lost.
Going the other direction works too. If you have been running step by step and want the agent to finish the rest, run /think --autopilot and it will pick up from where you left off.