Instructions file
Instructions for stuartshields/claude-setup, covering global protocol (v2026.2), 0. shortcuts & triggers, 1. mandatory workflow, 2. dynamic project initialization / migration and 3. agent routing.
Agent
Accessibility auditor for UI code. Use when reviewing components, pages, or user flows for WCAG 2.2 AA/AAA compliance — typically before shipping a UI feature. Covers semantic HTML, keyboard navigation, ARIA, forms, focus management, color contrast, and motion. Framework-aware (React, Vue, WordPress blocks, vanilla…
Agent
API contract reviewer. Use when changing REST, GraphQL, or RPC endpoints, modifying schemas, or before shipping API changes that downstream consumers depend on. Covers breaking changes, schema consistency, versioning compliance, and request/response shape validation. Read-only. Reports findings by severity with…
Agent
Supervisor agent for multi-file PR or pre-merge review. Spawns code-reviewer + security + perf in parallel via Task(), consolidates findings, drives fixes, and iterates until production-ready. Use for comprehensive audit-fix-review pipelines. For single-file or single-diff review, use code-reviewer directly instead.
Agent
Deep architectural research and recommendation agent. Analyzes codebase, researches approaches, and produces structured decision documents. Does NOT write implementation code. Use for technology choices, migration strategies, and design decisions.
Agent
Backend implementer. Use when building API routes, database schemas, server logic, or backend services that are clearly scoped and don't need design discussion first. Reads project CLAUDE.md, follows existing patterns. Designed for parallel execution alongside frontend-builder when work doesn't overlap. Returns the…
Agent
QA engineer with Chrome integration. Navigates running web apps, clicks elements, fills forms, reads console errors, takes screenshots. Use for interactive UI testing on localhost or deployed apps.
Agent
Meta-agent for Claude Code extensions. Use for: creating agents/skills/commands, debugging Claude Code behavior, MCP integration, hook configuration, prompt engineering for extensions, quality review.
Agent
Finds code that shouldn't exist — dead code, unused exports, orphaned files, stale dependencies, and accumulated cruft. Use for periodic hygiene or before major refactors. For code that exists and runs but is over-built or unnecessarily complex, use simplify instead.
Agent
General-purpose code reviewer for single files or focused diffs. Use when reviewing specific files or a single diff before merging. Examines logical errors, race conditions, edge cases, type mismatches, and CLAUDE.md compliance. Reports issues by severity with file:line references. For multi-file PR or pre-merge…
Agent
Pre-build feasibility checker. Use when you have a spec, plan, or requirements doc and want to confirm the codebase can support it before writing code. Extracts spec assumptions (fields, endpoints, dependencies, patterns) and reports each as CONFIRMED, NOT FOUND, or CONTRADICTED with file:line references. Read-only.
Agent
Implements a single frontend component, page, or feature with full fidelity. Reads project CLAUDE.md first, follows design system, outputs complete files. Designed for parallel execution - spawn multiple instances with isolation: worktree for independent components.
Agent
Background git operations agent - commits, PRs, branch management, release workflows. Runs on Sonnet to free main session. 3-tier safety system (read/safe-write/destructive).
Agent
Git history-aware reviewer. Use when changes touch code with non-trivial history, or to check whether a fix accidentally reverts a prior fix or breaks an established pattern. Uses git blame and log to surface why existing code looks the way it does. Read-only. Reports regressions, reverted fixes, and ignored context…
Agent
Database migration safety reviewer. Use before applying any SQL migration (D1, MySQL, PostgreSQL, SQLite), WordPress dbDelta, or ORM migration (Knex, Drizzle, Prisma) to production. Checks for destructive operations, backward compatibility, rollback plans, index coverage, data integrity, ordering, and performance…
Agent
Performance audit for web applications. Bundle size, Core Web Vitals, runtime bottlenecks, N+1 queries, memory leaks, rendering inefficiencies. Framework-aware (Next.js, Vite, Webpack). Use before deploys or when things feel slow.
Agent
Pull request description generator. Analyzes git diff and commit history, categorizes changes, generates structured PR with summary, testing checklist, and file impact table. Creates PR via gh CLI.
Agent
Fast haiku-powered agent for trivial single-file edits. Use for typo fixes, variable renames, small function additions, or config tweaks that are clearly scoped — typically under 50 lines. Hard guardrails prevent scope creep and context rot. Escalates to sonnet if the task turns out to be more complex than expected.…
Agent
Deep security audit adapted to the project's tech stack. Reads project config to tailor checks. Use before deploys, after adding auth/payment/user-input features, or for periodic review.
Agent
Test data generator. Analyzes database schema (Prisma, Drizzle, TypeORM), maps relationships, generates realistic seed data by environment (dev/test/staging). Writes seed files directly.
Agent
Analyzes live code for unnecessary complexity and suggests concrete simplifications. Use when code feels over-engineered, deeply nested, or harder to read than it should be. For finding code that's unused or dead (should not exist at all), use cleanup instead.
Agent
Test coverage reviewer. Use after writing or modifying production code, or when reviewing tests in a PR. Checks for missing test cases, untested branches, edge cases without coverage, and tests that don't actually assert anything. Read-only — does not modify code or write tests. Reports gaps by severity with file:line…
Agent
Writes thorough tests for existing code. Detects test framework from project config, reads CLAUDE.md first, follows existing test patterns. Handles unit, integration, and e2e tests. Designed for parallel execution with isolation: worktree.
Agent
Reviews frontend UI/UX for usability, accessibility, responsive design, and interaction quality. Use after building or changing UI components, pages, or user flows.