~ /Slash Commands

01 — index

Tested prompts for the jobs you repeat every day: /commit, /review, /fix-tests, /a11y. Built for frontend JS/TS projects. Each one detects your tooling, says what it may touch, and asks before anything risky. One Markdown file works in Cursor, Claude Code, Codex CLI, and Antigravity.

22 commands
4 agents
a safety tier on every command

Commit, Push & PR

Commit changes, push a feature branch, and open a GitHub pull request.

Create Pull Request

Push the current feature branch and open a GitHub pull request from existing commits.

Groom GitHub PR

Rebase, squash commits, and refresh a GitHub PR title and description to match the latest diff.

Resolve Merge Conflict

Three-way merge reasoning for Cursor's Resolve in Chat. Proposes each resolution before applying it.

Write Changelog

Draft release notes or a CHANGELOG entry from commits since the last tag, grouped by type. Text only.

Write Commit Message

Write a Conventional Commit message from the current diff. Text only — you commit.

Write PR Description

Generate or refresh a GitHub PR title and description from the branch diff — no rebase or squash.

YOLO Deploy

Commit and push on the current branch — fast path for personal projects.

Debug a Bug

Find the root cause of a bug before touching the fix — reproduce, hypothesize, test, and report. No guess-and-patch.

Fix Failing Tests

Run the project's JS/TS test suite, diagnose failures, and fix them until green.

Fix Lint Errors

Fix JS/TS linting and formatting issues without changing behavior.

Fix Type Errors

Run the TypeScript compiler and fix type errors properly — no any, no @ts-ignore, no non-null shortcuts.

Remove AI Code Slop

Strip AI slop from your branch — stray comments, needless try/catch, any-casts — and match the file's style.

Review Branch

Review your branch diff like a senior teammate — bugs, security, performance, then style, each with file:line. Report only.

Write Tests

Write tests for the selected code using your project's test runner and conventions. Plans the cases first.

Convert to TypeScript

Convert JavaScript to TypeScript with real types, matching your tsconfig strictness.

Modernize JavaScript

Rewrite code with modern JavaScript APIs from current Node.js LTS and baseline 2023 browsers.

Audit Dependencies

Run a security audit on your JS/TS dependencies and triage findings by real exposure. Suggests fixes, applies nothing.

Generate Favicons

Generate a complete favicon set and web manifest from one source image.

Remove Dead Code

Find unused files, exports, and dependencies with Knip, then remove only what you approve.

Update Dependencies

Update JS/TS dependencies within current semver ranges using the detected package manager.

Slash commands in a world of skills

Agents ship more skills every month: capabilities the model decides when to use. A slash command is the opposite. You pick the exact prompt, and it runs only when you type it.

You decide when

The model never guesses whether to run it. Type /commit and you get a commit message. Nothing fires in the background.

Same prompt, every agent

One Markdown file works in Cursor, Codex CLI, Antigravity, and Claude Code, whatever built-in skills each one ships.

Scoped, predictable output

Each command does one job and ends with a defined output: a commit message, a review with file:line findings, a list of what changed.

Every command says what it may touch

Each prompt has a Safety section that sets its limits. It asks before sensitive actions like committing, pushing, or overwriting files. The badge on each card shows the tier.

read-only

Reads your code and prints text. Never edits files or runs git writes.

scoped edit

Edits only the selection or paths you name. Asks first if scope is unclear.

broad edit

Plans first and asks before overwriting files or touching many of them.

git

Commits, pushes, or rewrites history. Running it is your approval.

Which git command do I need?

The git commands overlap on purpose: each fits a different moment between "I changed some files" and "it's merged". Find your situation and run that one.

I want a commit message, and I'll commit myself

/commitText only. Same idea as the sparkle button in Cursor's commit box.

Uncommitted changes on a team repo, and I want a PR

/commit-push-prBranches off the default branch if needed, commits, pushes, and opens the PR.

Already committed, no PR yet

/create-prPushes the branch and opens a PR from the existing commits.

PR is open, but the description is stale

/pr-descriptionDrafts a new title and body from the full diff. Applies only if you ask.

PR is open, and history is messy or behind the base branch

/groom-github-prRebases, squashes, and refreshes the PR after you approve the plan.

Personal project, just ship it

/yoloCommits and pushes on the current branch, even main. Not for shared repos.

A merge or rebase stopped with conflicts

/fix-merge-conflictsProposes each resolution as a diff before applying it.

Questions, answered

What is a slash command?

A Markdown file of instructions saved where your agent looks for commands. Type /name in chat and the agent runs that exact prompt, only when you ask. Nothing loads in the background.

Where does the command file go?

Cursor: .cursor/commands/ in your project (the install button does this for you). Antigravity: .agent/workflows/ in your project. Codex CLI: ~/.codex/prompts/, run as /prompts:name. Claude Code: ~/.claude/commands/, or .claude/commands/ in a project to share it with your team.

How do I pass arguments?

Type them after the command, like /favicon assets/logo.svg. Claude Code substitutes $1, $2, and $ARGUMENTS in the file. Other agents pass what you typed after the command along with the prompt, so commands here also ask for missing input.

Can I edit a command after installing it?

Yes. It's a plain Markdown file. Add your project's conventions, remove steps you don't need, or change the output format. Commit project-level commands so your whole team gets the same version.

Slash commands, skills, or rules: which should I use?

Use a slash command for a task you trigger on purpose: commit, review, fix tests. Use a skill for know-how the agent should apply on its own when a task calls for it, like React performance rules. Use rules (.cursor/rules, AGENTS.md, CLAUDE.md) for context that should apply to every request.

What do the safety tiers actually guarantee?

Each tier is a set of instructions the agent follows: what it may edit, and when it must stop and ask. They are not a sandbox. Keep your agent's own approval settings for terminal commands and file edits turned on, and review diffs before you commit.