ak skills kit rollback

Re-activate the previous revision the scope's skill lock records for a personal kit, from the skill store, offline.

Use ak skills kit rollback to project a personal kit's previous revision again. The command reads that revision from the skill store through the same transaction as ak skills kit apply; it needs no network and no foundation kits. The running revision becomes the previous one, so a second rollback returns to it.

Usage

ak skills kit rollback <name> [flags]

Options

Flag	Default	Description
`--dry-run`	`false`	Print the plan, its review findings and plan token; write nothing.
`--scope <scope>`	`project`	Lock and skills root scope: `project` or `global`.
`--runtime <list>`	The running revision's runtimes	`claude-code`, `codex`; repeatable. The running revision's runtimes are always included.

Shared flags are described in CLI conventions.

Requirements

The skill store must be on, after ak update --ssot or with AGENTKIT_SKILL_STORE=1, for dry runs too; with it off the command exits 2. The store gate follows --scope.

What it does

The command applies without prompting unless the plan needs review. In one journaled transaction under the kit lifecycle lock it writes the composed trees into the skill store under AGENTKIT_HOME, the projected files into each runtime's native skills root (mode 0644, never executable), and the scope's skill lock and receipts, owned by personal:<name>. It never rewrites the personal kit or the foundation kits, and never runs a script.

Being offline, it does not re-check whether a foundation version was revoked since; ak doctor reports yanked and revoked releases. A file you changed after it was projected is preserved and reported with ownership.modified.

Output

The text output has the shape of the ak skills kit apply output: the kit with its status and required approval, review findings, the new and previous revision, the plan token and the transaction plan.

JSON uses kind=skills.kit.rollback with the same data fields as skills.kit.apply:

personalKit, status, requiredApproval, planToken, findings[],
revision, previous, plan

Exit behavior

Exit	Meaning
`0`	Success, or nothing to change.
`1`	Command failure, such as a path too long or a recovery that is required.
`2`	The skill store disabled, or invalid flags or arguments.
`4`	Another AgentKit operation holds the kit lifecycle lock.
`5`	Personal kit or previous revision not found.
`6`	Review required, stale plan token, alias collision (including a legacy kit alias), or an unowned or modified path.
`7`	A link in a target path, or a digest mismatch.