Delegate a desktop or browser UI task to a Codex subagent (GPT 5.6 Sol by default) using Computer Use, where it reads the Mac screen and drives apps by clicking, typing, and scrolling. Use when the user wants Codex — or "GPT-5.5" — to operate a GUI app such as Chrome, Slack, Finder, Xcode, Figma, System Settings, or a website form, whether they say "have Codex click/open/fill/drag…" or type `/codex:computer`. Do NOT use for code or file work (use codex:implement to change code, codex:review to review it); prefer Claude's own Chrome tools for browser-only tasks when available. Flag risky real-world actions (sending messages, purchases, entering credentials) before invoking.
Delegate a coding task to a Codex subagent (GPT 5.6 Sol by default) that writes files directly in the working directory. Use when the user wants Codex — or "GPT-5.5" — to build, implement, write, add, refactor, fix, or otherwise change code, whether they say "have Codex do X", "get Codex to…", "hand this to Codex", or type `/codex:implement`. Do NOT use for read-only code review where Codex inspects a diff, branch, PR, or commit without editing (use codex:review); for driving desktop or browser UI (use codex:computer); when the user wants Claude to make the change itself rather than delegate; or for a trivial edit Claude can just do in one step.
Delegate a read-only code review to a Codex subagent (GPT 5.6 Sol by default) that inspects code and reports findings without editing anything. Use when the user wants Codex — or "GPT-5.5" — to review, look over, audit, or give a second opinion on a diff, branch, pull request, or commit, whether they say "have Codex review…", "get Codex's take on my changes", or type `/codex:review`. Do NOT use when the user wants Codex to fix or implement changes (use codex:implement); to drive desktop or browser UI (use codex:computer); or when they want Claude's own review rather than a Codex second opinion (use the code-review skill).
Run the full jj-to-PR workflow — describe changes, create per-commit bookmarks, push to origin, and open stacked pull requests with sensible defaults. This skill should be used when the user wants to "commit and open a PR", "push and create PRs", or ship a jj stack. Fills out a repo PR template (.github/PULL_REQUEST_TEMPLATE.md) when present; never adds Claude attribution or a Test plan section unless the template asks for it. Not for plain commits (use jj:commit) or rebasing (use jj:rebase).
Describe jj (Jujutsu) changes with conventional commit messages written as the user, never adding Claude attribution or co-authors. This skill should be used when committing session work in a jj repo — "commit this", "describe my changes", "jj commit", or when handed pre-gathered jj status/diff context. Honors a configured commit-message template when present. Not for git repos (use the git commit flow) and not for pushing or opening PRs (use jj:commit-push-pr).
Rebase the current jj (Jujutsu) changeset stack onto the latest trunk. This skill should be used when the user wants to "rebase my stack", "update onto main/trunk", or sync a jj stack with upstream. Stops and reports on conflicts rather than auto-resolving. For committing or opening PRs use jj:commit / jj:commit-push-pr instead.
Creates a new git/jj worktree for development work with proper setup, dependencies, and optional thoughts initialization. Use when user needs to create a new worktree for feature development or bug fixes.
Generates metadata for research documents and specifications including date/time, git commit, branch, and repository info. Use when creating research documents, handoffs, or any documentation that needs timestamp and git metadata.