cook-stack

Make a series of stacked PRs from the current work. Each PR in the chain should be able to land safely one at a time so that reviews and reverts are smoother. First, discuss with the user how to split the work up. Consult any repository-specific change size guidance applicable. Default to preferring very small PRs of <500 lines added. Use GitHub's Stacked PRs feature when creating the stack, then use `$writing-pr` for each one. Use the [`writing-pr`](../writing-pr/SKILL.md) skill for PR titles and descriptions. Have separate subagents monitor each PR in the stack. ## Stack ordering Where possible, split work into independently landable streams, and help the user understand the dependency graph. When there are legitimate inter-PR dependencies, try to put cosmetic/shim/prefactor work earliest so that semantic/functional changes can be reviewed as smaller diffs later in the stack. ## Events How to handle different situations while managing the PR stack. ### All PRs are created Update all PR bodies with cross-links to the rest of the stack. Bold the current PR's entry and append ⬅️ to it. ### Updates are pushed to a PR Notify subagents monitoring dependent PRs to use `$refresh-branch` and to push the results. ### PR is merged or closed Treat this as an update to the PR and also retire the subagent who was monitoring the merged/closed PR. If there are later PRs in the stack, `$refresh-branch` them.