Release to Production
The ikanban-release skill in your repository's .iKanban/ folder releases your project the software-factory way: one release commit, every part deployed in a fixed order with its proof, checks on production, then iKanban told. It uses your project's own CI and deploy commands, so it works for any stack and any cloud. Agent setup writes it with the full developer kit (Pro and Pro+), with .iKanban/prompts/release.md for agents without skills and .iKanban/runbooks/release.md for you.
release map ─► release commit ─► what prod lacks ─► "go" ─► parts in order (proof each) ─► prod checks ─► pipeline events ─► ship the release
The steps
| Step | What the agent does | You |
|---|---|---|
| Release map (once) | Reads your README, CI files and deploy scripts, and writes .iKanban/release-map.json: each part, its order, its deploy command, its proof and its rollback. | Check every command, then say "save it" and commit the file. |
| Release commit | Reads the commit staging runs. | — |
| What production lacks | Per part: the version production runs, the commits, pull requests, task keys and migrations it lacks. | Answer "go" to the table it shows. |
| Deploy | Each part in order from the release commit, then its proof. The proof must name the release commit; a green health check alone is not proof. | Sign in when a command needs your cloud sign-in. |
| Check production | Every part's version again, and one real request per change. | Sign in if a check needs it. |
| Tell iKanban | One ikanban pipeline event --stage release per part, for the release's task keys. Then get_release, and update_release status shipped when you say so. | Say "ship it". |
Gates
| Gate | Value | Where you see it | Change it |
|---|---|---|---|
| Production steps | only after your "go" | the question in the session | — |
| Pipeline events | Pro and Pro+ | 402 on Starter; the step is skipped | upgrade the plan |
| Releases on the board | every plan | — | — |
| The skill in your kit | full developer kit (Pro and Pro+) | Agent settings → Set up | upgrade the plan, or use the prompt |
When a step fails
The agent stops, shows the error and that part's rollback, and waits. Say "roll back" to run it, or fix the cause and say "continue": it goes on from the failed step with the same commit. It never drops data or rolls back a database migration as a fix.