dotfiles

Re-orient after a conversation compaction. Give a short, scannable summary of where the work stands and what's next. Use right after a compaction, or when the user asks "where are we?" / "what were we doing?".

Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.

Audits the copy of a multilingual project (next-intl, i18next, vue-i18n, Rails locales, etc.) for grammar errors, register inconsistencies (e.g. tu/vostè mixing in Catalan, tú/usted in Spanish), cross-locale structural gaps, non-idiomatic phrasing, and formatting issues. Discovers the locale files, presents findings in severity-grouped tables, then asks the user to confirm each group before applying any fix. Use when the user says "audit copy", "check translations", "find copy inconsistencies", "review locale files", or wants to check multilingual content quality.

Audit and realign a project's design system. Finds hardcoded colors/values that bypass DS tokens, unused DS tokens, and font inconsistencies. Use when the user wants to check DS coverage, hunt down hardcoded styling, or readjust styling to flow through the DS.

Capture before/after screenshots (or recordings) of a branch's UI/UX changes by running the base commit and the branch side by side against the same data, then saving paired shots to ~/Downloads. Use when the user asks to screenshot changes before and after, show a branch's visual diff, capture before/after of a PR, or document UI/UX improvements visually. Works for any stack — frontend, backend/API responses, CLI output, etc.

Commit pending changes by planning, staging, and creating separate atomic commits for unrelated WIP. Invoke automatically when the user asks to commit, stage changes, save work to git, or split WIP into commits.

Spawn a fresh subagent, in an unbiased context with no memory of the current conversation, to adversarially challenge the current plan, approach, or facts under discussion — then report back what holds up and what doesn't. Asks the user which model and reasoning effort to use for the challenger, recommending a more capable model than the current session. Use when the user wants a second opinion, wants to stress-test a plan with a sharper/independent reviewer, says "challenge this", "poke holes in this", "play devil's advocate", or wants an adversarial review.

Create, open, or draft a concise GitHub pull request (PR / MR) with gh pr create, following the repo's PR conventions and PR template.

How to fetch content from external tools when the user pastes a URL. Invoke automatically whenever a message contains a URL pointing to linear.app, github.com, gitlab (any host), notion.so, or sentry.io — before attempting WebFetch on those domains.

List the open GitHub issues for the current repository using the `gh` CLI. Use when the user asks to see, list, or check open issues, the issue backlog, or what's pending on GitHub.

Fetch open GitHub issues for the current repo, group them by conflict risk, then implement each group in parallel via Sonnet subagents in separate git worktrees. Each worktree builds and starts the app on its own port, and the skill returns a comparison table for manual QA.

Rules for Next.js and design-system work. Invoke automatically when editing Next.js code (app/, pages/, route handlers, server/client components, `"use client"`, server actions), when changing styling/colors/borders/backgrounds, when working with design tokens, or when the user mentions ds, design system, tokens, theme, or Tailwind/CSS variables.

Rules for writing, running, and fixing tests. Invoke automatically whenever the user asks to write a test, add a test, fix a test, run tests, debug a failing test, check test coverage, or whenever the task involves touching files matching *.test.*, *.spec.*, __tests__/, or test/ directories.

Create new agent skills with proper structure, progressive disclosure, and bundled resources. Use when user wants to create, write, or build a new skill.

Run a comprehensive accessibility audit on a web project using axe-core (runtime browser scanning) and/or eslint-plugin-jsx-a11y (static analysis). Use when the user says "run accessibility scan", "a11y audit", "check accessibility", "WCAG compliance", or invokes /a11y or /accessibility.

Compact the current conversation into a handoff document for another agent to pick up.

Implement a task in an isolated git worktree, install dependencies, link .env files, then start the dev server on a free port and report the URL. Use when the user wants to run changes in parallel with other instances on the same repo without disturbing the current working tree.

Copy the last plan from the conversation into the clipboard. Use when the user asks to copy a plan, save a plan to clipboard, or wants to share a plan.

Send a push notification to the user's phone via ntfy.sh. Use proactively after completing any long-running task (builds, tests, deployments, large refactors), when the user explicitly asks to be notified, or when something requires immediate attention. Do not use for routine short interactions.