dotfiles

Use when a tool, library, or framework doesn't support what you need and official docs don't cover it — finding solutions off the happy path. Triggers on "how do I make X do Y", "X doesn't support Y", undocumented behavior, bleeding-edge tool limitations, workarounds for missing features, or when the mainstream answer is "you can't do that"

Iteratively review, refactor, verify, and commit toward less code by sharing primitives and improving functional composition. Use when the user asks for a strict refactor review loop, "get a review strictly for refactoring to less code", "repeat review/fix/commit until empty", or similar wording.

Use when the user says "sync" or wants to commit, rebase, push, and converge this dotfiles repo. Uses just up as the public converge command.

Poll, diagnose, and report on a GitHub PR's CI. Use this skill whenever you're about to watch a PR's CI to a terminal state — after pushing commits to a PR, after /git_sync, when the user says "is CI green", "wait for CI", "check the PR's checks", "monitor CI", "did CI pass", "watch the workflow", "fix CI failures", or any phrasing that asks whether a PR is ready to merge based on its checks. ALSO use proactively whenever you push a commit to a branch that has an open PR — never spin up an ad-hoc `gh pr checks --watch` or count-based poll. This skill is the only authority for PR CI status; ignore any older guidance suggesting otherwise.

Complete GitHub PR review-comment closeout. Use whenever the user asks to address, fix, handle, clear, close, resolve, respond to, or follow up on PR review comments, review threads, requested changes, bot findings, CodeRabbit, Greptile, or Devin review feedback. Addressing a review thread includes a written thread reply and resolving the thread after the fix or justification is pushed.

Orchestrate land-pr, gh-ci, and land-merged into one end-to-end landing pipeline. Use when the user asks to land complete, land and merge, merge on green, finish a PR end to end, or land without another confirmation loop.

Use after a GitHub pull request has already been merged to refresh the local develop worktree, remove the local git worktree that backed the merged PR, and delete the merged local branch. Trigger on requests like "land merged", "cleanup merged PR worktree", "sync develop and delete the PR worktree", or "PR is merged, clean up".

Prepare a branch or GitHub PR to an open, ready PR by composing land-worktree first, then creating a PR when missing, moving it out of draft, scanning existing Greptile review feedback, addressing actionable Greptile feedback when present, watching PR checks, and triggering E2E Local Sign Off after other checks are green when required. Use when the user says to land to PR, prepare a PR, create or ready the PR, undraft and handle existing Greptile feedback, finish review-comment closeout, handle local sign off, or get the PR green without merging.

Prepare the current git worktree and branch for landing by resolving task context, creating a task branch from a detached base worktree when needed, committing intended dirty work when needed, and pushing the branch upstream. Use before land-pr when the worktree may be detached, unnamed, dirty, missing an upstream, or otherwise not ready for PR creation.

Use Codex loop or heartbeat polling plus a source scanner to treat `@agent` comments as chat messages from the user. Use when the user asks to enter comment chat mode, permanently poll agent comments, watch a file for `@agent` notes, resolve inline agent comments, or keep responding to code comments until they are deleted.

Turn off Heartbeat PR E2E mode by editing the generated PR Dashboard comment checkbox and syncing the E2E Local Sign Off status. Use when a PR's broad E2E gate is unnecessary for the current change, when the user says to turn E2E off, disable E2E, remove E2E local signoff, or when the PR dashboard has Local or Local Sign Off checked but the targeted CI checks already cover the work.

The authoritative playbook for driving a browser with the Claude in Chrome integration. Use whenever browser work is starting — web app testing, navigating or clicking pages, reading the DOM or console, filling forms, extracting data, recording a demo GIF, or any mcp__claude-in-chrome__* tool — and whenever the user mentions Chrome, Edge, the browser extension, "the chrome plugin", remote debugging, or browser permissions, even if no specific tool is named. Load it BEFORE touching any browser tool; it prevents conflating the extension with CDP / remote-debugging and guessing at the tooling.

Use when the user asks "what is this", "explain this", "why/how/when does this code run", "help me understand this block", or points at suspicious or wrong-looking code and needs a fast system-level map for refactor decisions. Produces before/after reference-site reports with code evidence and plain English bullets, not local-only explanations.

Run a deep Codex (OpenAI) headless research pass using gpt-5.4 with full-auto mode. Use this skill whenever the user asks for "codex research", "openai research", "gpt research", "get a second opinion on the theory", or /codex-research. Sends a research question to OpenAI's Codex CLI for cross-model literature and algorithm research. Produces a written research report.

Use when the user says "$defog", "/defog", "plain English", "too dense", "translate that", "what does that actually mean", "human version", "less jargon", "give me the TLDR", "anti-slop funnel", or otherwise asks to turn a wall of agent text into something clear, short, and easy to read.

Source-first workflow for building, debugging, reviewing, or modifying privileged high-trust local development Chrome extensions, especially Manifest V3 service workers, native messaging hosts, extension popup/options pages, browser profiles, Chrome launch behavior, TypeScript/Effect extension architecture, or extension install/reload flows. Use when work touches chrome-extension:// URLs, manifest.json, chrome.runtime/chrome.* APIs, nativeMessaging, service workers, extension storage, profile/native-host diagnostics, or local devtool browser integrations.

Dispatch work to parallel Claude Code sessions via cmux workspaces. Use this skill for "$dispatch-claude" or whenever you need to hand off one or more independent tasks to separate Claude instances — for example, when acting as a dispatcher and the user asks you to "run these prompts", "fan out", "spawn sessions", "dispatch Claude", or when you've prepared prompts that should each run in their own Claude session. This is agent-invokeable only — the agent uses it to launch Claude work, not the user directly. Always prefer this over asking the user to copy-paste prompts into new terminals.

Dispatch a peer Codex agent. Use for "$dispatch-codex", "peer Codex agent", "peer agent", "separate Codex agent", "new Codex thread", or when the user asks for another Codex agent working beside this one. This is not a subagent.

Dispatch an in-thread Codex subagent. Use only for "$dispatch-codex-sub", "subagent", "sub-agent", "in-thread helper", or "multi-agent subtask". Do not use for "peer agent".

Read user-authored review comments via `dunk comments` (or `.dunk/comments.json` for the active repo) and address each one. The TUI is for humans; agents drive the review through the CLI.

Find and land one high-leverage Heartbeat app dependency removal or update. Use when the user asks to audit app dependencies, remove weak or unused dependencies, update stale packages, make dependency usage Effect-native, create an internal monorepo package for a dependency concern, or run a repeatable dependency gardening pass that ends with land-pr when a real change is made.

Use when the user says "/loop", "loop this", "start loop", "enable loop mode", "keep going", "don't stop", "continue until done", "run unattended", "cancel loop", "disable loop", "is loop on", or "what is the escape token". This skill manages a stop-hook continuation loop for the current Codex session only.

Open the Heartbeat local dev app in the user's Chrome browser at a requested app view. Use when the user says "meet me in...", "take me to...", "open the app at...", "get a Chrome tab going", "meet in Page Editor", or otherwise asks Codex to autonomously start or align the dev stack, connect the Chrome plugin, and navigate to a local app place.

Force the agent to own errors instead of dismissing them as 'pre-existing'. Use /own-it when the agent claims something is 'pre-existing', 'not from this PR', 'already on develop', 'out of scope', 'acceptable for this PR', or any other framing that avoids responsibility for a problem visible on the current branch. Also trigger when the agent proposes deferring a fix with FIXME/TODO, suggests an error is 'not introduced by this work', or recommends 'leaving it' because trunk has the same issue. The agent's job is to leave the branch clean — trunk's problems are irrelevant if they're visible here.

Use when Codex needs the user's secrets, Apple Passwords.app, macOS Keychain, or saved 2FA/TOTP verification codes for credential reads/writes without exposing secrets. Prefer the `security` CLI credential cache first for recurring agent username/password logins, falling back to Passwords.app on cache miss, verification/update, or MFA code prompts. Triggers include "$passwords", "$secrets", "secrets skill", "use Passwords", "open my pw app", "copy password from Passwords", "copy 2FA code from Passwords", "save/update this password", "use my security cli", "unlock Passwords", or any login/workflow where credentials live in Apple Passwords.

Open Plannotator on the latest rendered assistant message and use the returned annotations to revise that message or continue.

Turn an idea or objective into a goal package for /goal. Interviews the user, builds a reviewed fact sheet via Plannotator, then explores the codebase to produce an execution plan.

Mode switch that turns the agent into a relentless technical coach and pair-designer. Stops all summarizing, praising, hedging, and chat. Starts reading source code, finding footguns, proposing 10x abstractions, and being blunt. Use when the user says "/push-me", "push me", "coach mode", "stop being nice", "be critical", "challenge me", "stop chatting", "designer mode", "architect mode", "tear it apart", "what am I missing", or expresses frustration with shallow/safe/diplomatic responses.

Use when the user says "/readme", "write a readme", "improve the readme", "update the readme", "readme for this project", "contributing guide", "write contributing", or when a new package needs its README. Also triggered by natural language like "document the new feature in the readme", "refresh the readme", or "update the contributing guide".

Reset a long or messy conversation back to one coherent current view. Use when the user says "/return", "return to the main thread", "where are we", "what are we actually doing", "level set", "reset me", "current truth", or otherwise needs the thread collapsed into what is done, what is not done, what matters now, and the next move.

Use when Codex needs to log into Revenu Quebec My Account with the Government Authentication Service, including Chrome navigation, Passwords.app credential handoff, Mail.app 2FA/security-notice handling, and post-login cleanup of recent RQ auth emails. Triggers include "rq login", "log into my rq", "Revenu Quebec gov auth", "Government Authentication Service", or RQ document/download flows that require an authenticated session.

Use for high-investment purchase decisions where the user wants to research a category, evaluate options, and shop with confidence. Triggers on phrases like "I'm shopping for X", "help me find a Y", "research X to buy", "compare options for Z", "I want to buy a quality A", "find me a [category]". Especially when aesthetic, material spec, or vendor preferences matter (home goods, clothing, furniture, appliances, gear). Coordinates a research-dispatch + user-visual-verification loop, scaffolds a matter directory with a living vendor catalog and locked spec, and closes cleanly when purchases happen. ALWAYS use this skill the moment the user signals they're shopping for something non-trivial — even before asking clarifying questions.

Use when the user says "start" with a Linear issue, asks to start work on a Linear ticket, or wants a new Codex peer agent launched in the current Codex project, usually with Codex App's first-class worktree mode, to work toward a draft PR.

Run the framework-agnostic visual oracle ratchet algorithm for screenshot, golden image, and other image regression tests. Use when deciding whether a visual oracle can ratchet forward, reviewing visual diffs, handling image baseline failures, or designing adapter skills for Vitest, Playwright, or similar tools. Enforces human approval before any baseline overwrite.

Run the visual-oracle-ratchet workflow specifically for Vitest Browser screenshot tests using toMatchScreenshot. Use when Vitest Browser screenshot tests fail, when updating Vitest screenshot baselines, when reviewing Vitest actual/diff artifacts, or before running Vitest with -u for screenshots.

Sync the current branch with the main development branch using merge commits. Use this skill whenever the user says "sync", "sync with develop", "update branch", "merge develop", "resolve conflicts", "PR has conflicts", or /git:sync. Also trigger when the user mentions a PR needing updates, a branch being behind, or merge conflicts on a pull request. This skill handles worktree detection, fetching, merging, checks, and pushing — the full PR sync workflow. Never rebase; always merge.

Use when authoring, debugging, reviewing, or extending a Vite plugin — or a Rollup/Rolldown plugin used with Vite. Covers the plugin object, hook selection (resolveId/load/transform, config/configResolved/configureServer/transformIndexHtml/handleHotUpdate), virtual modules, dev-server vs build behavior, plugin ordering, apply/enforce, and packaging a plugin.

Use when installing, updating, evaluating, or debugging Claude Code plugins, configuring or troubleshooting MCP servers, managing permissions or allow-deny rules, browsing or publishing to marketplaces, advancing CC plugin or MCP setup, or diagnosing any plugin, MCP, permission, or marketplace failure.

Use when working with GitHub PRs — resolving review comments, checking CI status, writing PR comments, triggering AI re-reviews, or running PR review integration tests.

Run QA criteria against the current issue's codebase and produce structured findings. Use when asked to review, audit, or QA the current issue. Triggers on /issue:review.