Right thing?
An interview turns the request into an agreed brief. You approve what success means, and what you are explicitly not building.
Gate 1A crew for your next Claude Code project
Turn an idea into work you can ship.
Capstan coordinates the planning, building, independent review, and verification. You approve the concept, the plan, and the delivery.
Free plugin · MIT licensed · Your Claude usage applies
/capstan:effort add rate limiting to the public API You talk to the Architect. It interviews, plans, and never builds or reviews. The run stops at each gate. You decide when it continues.
The process
One feature, through the process.
The run ends at each gate. You decide when it continues.
Right shape?
The plan is cut into slices: what runs in parallel, what was assumed, and where each slice stops.
Gate 2Ready to deliver?
Each Builder works one slice, test first. A fresh Reviewer checks standards and spec independently, then the repository's own checks run.
Gate 3Leave with the reasons.
The Courier packages the output and writes a permanent note. You commit it once review passes.
After approvalThe difference
Separate jobs. Separate judgment.
The builder does not mark its own work.
ScoutArchitectBuilderReviewerCourier
Agree before the build
The run ends at every gate. You decide when it continues, while a change is still a decision on a page.
Separate judgment
The Builder does not review its own slice. A fresh Reviewer checks two axes, and nothing polls or runs unprompted.
Keep the reasons
Decisions and effort state survive the session, so the next run has a claim file to resume from.
BytesNation product: agent methodology from concept to delivery. capstan.bytesnation.com · GitHub · Published by BytesNation