Stack
An open-source VS Code extension for running AI coding agents such as Claude Code and Codex CLI in parallel within isolated git worktrees, while retaining the IDE's code navigation and debugging capabilities.
Timeline (last 90 days)
The real reason Stack is popular is that it proves engineering judgment can be compressed into Markdown, distributed, and executed; competition in AI programming is shifting from whose model is stronger to whose workflow is more solid.
An internal plugin called Stack within the Cursor team—the author's posted internal leaderboard shows his skills were invoked by colleagues about 9,000 times in a week.
Stack is an MIT-licensed open-source plugin made up of a bunch of Markdown files plus a few scripts; it is not a new model and not a paid SaaS.
Stack's content is divided into four categories and more than twenty workflow skills: 23 task Playbooks, 24 engineering principles (one per file), and two dedicated sub-Agents.
Stack requires remembering only one command, "slash Potato Gun Mode"; upon receiving a task it automatically matches the corresponding Playbook, copies the script steps verbatim into the task list, then invokes other skills step by step and dispatches the work to sub-Agents.
In Stack's default configuration, writing code, feature development, refactoring, and bug fixing are assigned to Grok; decisions requiring judgment and writing documentation go to Opus; for code review, the Interrogate skill has models from different families review the same Diff simultaneously.
Stack's Autopilot script can automatically verify and advance a string of PRs one by one, with a review panel giving a verdict each round.
Stack's "Lazy Protocol" principle advocates leaning toward deletion, solving problems with the smallest change, subtracting before adding, and cleaning up redundant validators and dead code before starting work.
Stack's "Attack the Premise" principle stipulates: when two fixes sharing the same premise both fail, write the premise down and test the premise directly.
The entire Stack plugin has no skill that does planning; the author does not believe in planning before acting, and holds that the best specification is the code itself.
13 more timeline events
All Facts (20)
The real reason Stack is popular is that it proves engineering judgment can be compressed into Markdown, distributed, and executed; competition in AI programming is shifting from whose model is stronger to whose workflow is more solid.
50%UnverifiedAn internal plugin called Stack within the Cursor team—the author's posted internal leaderboard shows his skills were invoked by colleagues about 9,000 times in a week.
50%UnverifiedStack is an MIT-licensed open-source plugin made up of a bunch of Markdown files plus a few scripts; it is not a new model and not a paid SaaS.
50%UnverifiedStack's content is divided into four categories and more than twenty workflow skills: 23 task Playbooks, 24 engineering principles (one per file), and two dedicated sub-Agents.
50%UnverifiedStack requires remembering only one command, "slash Potato Gun Mode"; upon receiving a task it automatically matches the corresponding Playbook, copies the script steps verbatim into the task list, then invokes other skills step by step and dispatches the work to sub-Agents.
50%UnverifiedIn Stack's default configuration, writing code, feature development, refactoring, and bug fixing are assigned to Grok; decisions requiring judgment and writing documentation go to Opus; for code review, the Interrogate skill has models from different families review the same Diff simultaneously.
50%UnverifiedStack's Autopilot script can automatically verify and advance a string of PRs one by one, with a review panel giving a verdict each round.
50%UnverifiedStack's "Lazy Protocol" principle advocates leaning toward deletion, solving problems with the smallest change, subtracting before adding, and cleaning up redundant validators and dead code before starting work.
50%UnverifiedStack's "Attack the Premise" principle stipulates: when two fixes sharing the same premise both fail, write the premise down and test the premise directly.
50%UnverifiedThe entire Stack plugin has no skill that does planning; the author does not believe in planning before acting, and holds that the best specification is the code itself.
50%UnverifiedStack's principles hold that the Agent should not stop to ask people at every turn: keep working, present the results afterward for people to correct, and only irreversible operations require confirmation.
50%UnverifiedStack's author turned his own taste into a plugin and also gave you the method for making plugins—the plugin ships with an "Ultimate Me" skill that can go through your recent chat history with the Agent, analyze your actual working habits, and generate a dedicated Mode named after you.
50%UnverifiedThe Stack company (Stacker) combined tar's container idea with LZ compression to launch Stacker, which first compresses files and then packs the compressed data into a virtual disk container, transparent to the user, effectively doubling disk capacity.
50%UnverifiedStack currently supports macOS and Linux and depends on tmux; on Windows it is limited to WSL, and native notifications are limited to macOS.
50%UnverifiedStack is not yet listed on the VS Code Marketplace, its current version is 0.1.0, and it must be installed via a .vsix file.
50%UnverifiedWhen Stack registers hooks in Claude Code, it writes to ~/.claude/settings.json and keeps a .stack-backup backup before the first modification.
50%UnverifiedIn Stack, Codex's hook is an experimental feature and is disabled by default; it must be manually enabled and the hook trusted, otherwise its sessions will not change state in the sidebar.
50%UnverifiedStack provides each worktree with an independent debug port and launch.json, with the default debug configuration being Node, which can be changed to any debugger supported by VS Code.
50%UnverifiedStack has a built-in MCP server, and via the /stack-task instruction it can work with Jira MCP to read ticket content and create tasks.
50%UnverifiedThe Stack project currently has 2102 tests, achieving 100% coverage of the measured code, with the React view as the interface excluded from measurement.
50%