arndvs

mobile-dev

Expo mobile development workflow — clean state reset, startup sequence, native rebuild decisions, and Android emulator/QEMU troubleshooting for Launch Core and its forks (Foreword, Scorpion, Aligned, Cast).

architect

Comprehensive implementation planning with vertical slices. Use when asked to 'act as an Architect', 'plan this', 'create an implementation plan', 'slice this into tasks', or when a task needs decomposition before execution.

Code & Development Listed

code-review

Focused review of staged or recent changes to find edge cases, logic errors, and integration risks before merging. Use when asked to 'do a code review', 'review my changes', 'review this PR', 'check my diff', 'review staged changes', or 'pre-merge review'.

codebase-audit

Ruthless codebase audit reporting only real problems. Use when asked to 'audit', 'code audit', 'codebase audit', 'review code', 'find bugs', or 'code review'.

do-work

Core execution loop for implementing tasks. Use when asked to 'do work', 'implement', 'build this', 'fix this', 'plan execute clear', 'plan then execute', 'work loop', or when working through a plan or backlog item.

document

Write, update, or audit documentation. Use when asked to 'document this', 'write docs', 'update the README', 'add JSDoc', 'write a changelog', or 'create an ADR'.

error-audit

Use when repeated Claude Code session errors need analysis; scans session transcripts for errors, clusters by root cause, and suggests remediations.

explore

Deep codebase exploration using parallel subagents. Use when asked to 'explore', 'understand', 'investigate', 'map out', 'how does X work', or 'audit' a part of the codebase.

Web & Frontend Listed

frontend-component-style

Frontend component file structure, naming, and layer separation for new or refactored components.

grill-with-docs

Use when stress-testing a plan against the project's domain model — grills the design, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as decisions crystallise.

halbert-copy-editor

Edit and punch up sales copy, marketing emails, landing pages, ads, and any persuasive writing using the Halbert Copywriting Method.

npm-security-audit

Layered security audit on npm/pnpm/yarn projects before installing or running. Use to vet a GitHub repo, npm package, or local project before npm install/start/npx.

performance-audit

Deep system and codebase performance audit — memory leaks, process bloat, network latency, and runtime profiling. Use when asked to 'performance audit', 'why is everything slow', 'find memory leaks', 'check system health', 'diagnose slowness', or 'perf check'.

plan-archive

Use after merging a PR or during periodic cleanup to archive plan-mode files by linking them to merged PRs.

prd-to-issues

Break a PRD into independently grabbable GitHub issues with dependency relationships. Use when asked to 'break this PRD into issues', 'create issues from PRD', 'plan the work', 'create a kanban', or after writing a PRD to prepare work for execution.

press1-check

Use when Claude Code sessions had many manual approval ("press 1") prompts or when auditing hook permissions; identifies which Bash commands required approval.

research

Cache expensive exploration into a research document before building. Use when asked to 'research', 'investigate before building', 'gather context', 'flush unknowns', or before a large feature where unknowns need flushing.

review-pr-copilot

Address GitHub Copilot review comments on the active PR by triaging into confidence tiers, fixing in atomic commits, resolving threads, and re-requesting review.

sanity-best-practices

Sanity development best practices — schemas, GROQ, TypeGen, Visual Editing, Functions, Blueprints, and framework integrations. Use when working with Sanity content, schemas, queries, or Studio.

session-close

Pre-flight checklist that runs quality gates before ending a coding session.

sketch-the-solution

7-phase UX design process from user stories to tested interfaces. Use when asked to 'sketch the solution', 'design UX', 'UX process', 'product design process', or when going from idea to interface design.

skill-scaffolder

Meta-skill for creating new agent skills that involve multi-step automation, browser navigation, state tracking, evidence capture, and both local (VS Code Insiders) and VPS (Playwright) execution. Use when the user wants to 'create a skill', 'build a new skill', 'scaffold a skill', 'make a skill for X', or describes a multi-step agentic workflow they want to automate.

stress-test

Invoke to run the adversarial stress test protocol against ctrl+shft rule compliance. Guides a human reviewer through systematic attempts to break rule boundaries — the same methodology a skeptical senior engineer would use before trusting the system in production.

systematic-debugging

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.

visual-feedback

Use when implementing UI, checking dark/light mode, or validating animations — adds a visual feedback loop via browser screenshots so frontend changes are verified, not assumed.

write-a-prd

Write a Product Requirements Document from a rough idea. Use when asked to 'write a PRD', 'create a PRD', 'plan a feature', or when starting a new feature that needs scoping.

improve-architecture

Explore the codebase for architectural improvements using deep module analysis. Use when asked to 'improve architecture', 'improve codebase', 'find shallow modules', 'refactor architecture', or for periodic codebase health checks.

tdd

Red-green refactor workflow for test-driven development. Use when asked to 'write tests first', 'TDD', 'red-green refactor', 'test-driven', or 'failing test first'. Backend-only — do not use for frontend components.

atomic-commits

Atomic commits on a feature branch with conventional messages — Commit mode for checkpoints, Ship mode for PR. Use when committing, checkpointing, shipping, pushing, or creating a PR.

pr-preflight

Exhaustive pre-PR audit that front-runs code review tools by catching the same issues Copilot/reviewers find iteratively, across any stack.

caveman

Ultra-compressed conversational replies — drops articles, filler, and hedging to cut output tokens while keeping full technical accuracy. Deliverables (plans, PRDs, issues, docs, commits, code) always stay normal prose; only the surrounding narration compresses. Use when asked to 'talk like caveman', 'caveman mode', 'be terse', 'fewer tokens', 'compress your replies', or invokes /caveman.

Data & Documents Listed

data-object-conventions

Enforce consistent structure, formatting, and placement for static data objects (catalogs, config, content, lookup tables, enums-as-data) in TypeScript/JavaScript codebases. Use whenever creating, editing, refactoring, or reviewing a data file or data object — including extracting data out of logic files, splitting a monolithic file, deciding where data should live, or formatting a data array. Triggers on phrases like 'data file', 'data object', 'catalog', 'lookup table', 'extract the data', 'split this file', 'format this data', 'where should this data live', 'this data is ugly'. Do NOT use for runtime state, API responses, or database rows — those are data-in-motion, not static data.

Web & Frontend Listed

repo-hygiene

Drives a long-running, incremental refactor of a messy codebase toward reusable layouts, extracted components, and clean structure. Use whenever the task involves auditing a repo for structural debt, planning the next step of a multi-week refactor, or generating a single well-scoped backlog issue for automated/nightly refactor runs. Also use it when asked to clean up huge monolithic pages, extract layouts or components, migrate page CSS to Tailwind, set up a repeating refactor workflow, or 'improve repo hygiene'. Stack-aware: dispatches to Astro, static-HTML, Worker-TS, or Python audits automatically. Trigger even if the user only says 'what should I clean up next' about a project.

agent-assets

Audit the agentic infrastructure committed to a repo — reusable commands, autonomous workflows, vendored documentation, and tool or MCP configuration — and propose what to add, consolidate, or retire. Use when the same prompt is being retyped across sessions, when agent workflows exist but nobody can tell whether they ran or what they did, when orientation files have gone stale or contradict the code, when agents repeatedly fetch the same external docs, or when planning to move work from interactive prompting to scheduled autonomous runs. Also use for requests like "what should be a slash command", "review our .claude directory", "why do our agent workflows keep silently failing", or "audit our AGENTS.md".

agent-navigability

Audit a codebase for how easily a coding agent can navigate it, and propose structural fixes. Use this whenever an agent keeps re-exploring the same code, burns context finding where things live, edits the wrong file, or misses call sites it should have found; when reviewing structure, naming, module boundaries, or type usage for legibility rather than correctness; when onboarding cost is high for agents or new engineers; or when someone asks why agent runs on this repo are slow, expensive, or unreliable. Also use for requests like "restructure this for agents", "why does it keep missing files", "review our naming", or planning a refactor whose goal is comprehension rather than behaviour. Includes a migration mode for repo-wide restructuring such as separating client from server, splitting a monolith, or changing the top-level layout, which produces a phased plan rather than a findings list.

agent-session-friction

Analyse a completed agent session transcript for evidence that the codebase made the work harder than it needed to be, and record those observations for later consolidation. Use at the end of an agent run, when reviewing a transcript to find what slowed an agent down, when deciding which repo problems are actually costing something, or when someone asks why a run took so many turns or burned so much context. Also use for requests like "audit this session", "what friction did this run hit", "post-run review", or when building automation that collects friction evidence across many runs. Produces observations with measured cost, not issues — publishing is a separate consolidation step.

agent-testability

Audit whether a codebase gives a coding agent a reliable way to verify its own work, and propose fixes. Use whenever an agent reports success but the change is broken, when tests exist but don't catch regressions, when the feedback loop is too slow or too noisy for an agent to use, when test failures don't say what broke, or when someone asks how to make a repo safe for autonomous or unattended agent runs. Also use for requests like "why does it keep saying it's done when it isn't", "review our test setup for agents", "what should agents run before opening a PR", or planning a testing strategy where the consumer is an agent rather than a human.

Code & Development Listed

logging-audit

Audit code for logging and observability quality, retrofit code onto whatever logger a repo already uses, and diagnose why a failed run left no usable evidence. Use this whenever reviewing a diff or PR for log quality, cleaning up or standardising logging in a module, judging whether error handling would actually be diagnosable in production, deciding what a catch block should do, or running a postmortem on a job or agent run that failed without explaining itself. Works in any language and any repo. Trigger it for requests like "review the logging here", "why was this run impossible to debug", "is this catch block okay", "make this failure easier to trace", or any investigation of a job that died silently — even when the user never says the word "logging".

grill-me

Interview the user relentlessly about a plan or design until reaching shared understanding. Use when asked to 'grill me', 'interview me', 'ask me questions about this', or before writing a PRD to flesh out vague ideas.