mvp-builder

Drive Xcode and the iOS/macOS toolchain through two complementary surfaces: the live-IDE bridge (xcode-cli + MCP tools, talks to an open Xcode workspace via SourceKit and AppleScript) and the standalone CLI (xcodebuild + xcrun simctl, runs headless). Use whenever the user wants to build, run, or test an Xcode project, see compiler errors, fix Swift code, render a SwiftUI preview, look up an Apple API, manage iOS simulators (boot, install, screenshot, log capture), send push notifications or deep links to a simulator, build a release archive, or do anything that needs Xcode's project context or the iOS Simulator. Trigger on direct mentions — Xcode, xcode-cli, xcodebuild, xcrun, simctl, SwiftUI, iOS app, macOS app, Swift Package, .xcodeproj, .xcworkspace, scheme, simulator, UDID, #Preview — and on indirect cues: "build my app", "what's broken", "fix the build", "run tests in MyTarget", "render this SwiftUI view", "boot a simulator", "install on iPhone 16 Pro", "take a screenshot of the simulator", "stream logs

Structured reasoning harness for problems where shallow thinking fails or hallucination is costly. Use whenever a question requires multi-step reasoning, involves competing hypotheses, has multiple interacting components, or carries irreversible consequences. Activate for debugging with unclear root causes, intermittent or multi-component issues, incident post-mortems, choosing between multiple valid approaches, architecture or design tradeoffs, causal chain analysis, high-stakes decisions, and any task where the first plausible answer is suspiciously easy. Provides reasoning modes (abductive, counterfactual, first-principles, inversion, and more), grounding techniques against confabulation, adversarial self-checks, root-cause playbooks, and automatic multi-approach synthesis (independent angles run in parallel and compared) — all through the `mcp__sequential-thinking__sequentialthinking` tool. Skip for trivial lookups, obvious fixes, and single-step questions.

Decision-tree framework for architectural trade-offs during implementation planning. Classifies the subject into categories (data-heavy, real-time, offline-critical, media-heavy, integration-heavy, frequent UI iteration), loads the minimum relevant references, supplies a question bank with opinionated defaults for caller-driven dialogue, and returns structured decisions (Architectural Decisions, Required Behaviors). Use whenever architectural choices must be committed to a document — protocol selection, pagination or caching strategy, offline sync approach, real-time transport, media upload strategy. Produces decisions, not implementation rules. Implementation details belong in code-level rules files.

Comprehensive codebase analysis for building mental model of project structure, dependencies, and implementation context. Use when needing to: (1) Understand project architecture before review or documentation, (2) Find dependencies and shared modules, (3) Trace execution paths and abstraction layers for similar features, (4) Locate implementation markers (AICODE-*), (5) Prepare context for review, memory generation, or agent creation. Triggers on: analyze code, load code context, scan codebase, understand project structure, trace feature.

Extract design tokens, styles, component metadata, and screen structure from Figma files via MCP. Use whenever a figma.com URL appears, when someone mentions Figma tokens, variables, design system extraction, screen captures, text styles, shadows, component properties, or library assets. Trigger for phrases like 'get from Figma', 'extract design', 'Figma variables', 'design tokens', 'capture screens', 'get typography from Figma', 'extract shadows', 'find components in library', 'get code syntax for tokens', or any figma.com/design link. Even if the user just pastes a Figma link without instructions, use this skill to parse it and decide what to extract. Covers three extraction depths: Level 1 (quick read-only MCP), Level 2 (deep extraction via Plugin API scripts), Level 3 (library search + component property metadata).

Visual QA gate for frontend code. ALWAYS use before delivering any UI changes - navigate, screenshot, verify console is error-free. Use for: responsive testing across viewports, form/interaction testing, debugging hydration and render issues. Catches visual bugs before users see them.

AI-native code commenting system with grep-searchable AICODE-* markers for cross-session memory. Use when: working with code files, leaving notes for future sessions, breaking down complex tasks, documenting non-obvious logic, recording bug fixes. Triggers: start coding session, edit code, complex logic, leave note, todo for later, debug session, bug fix, AICODE markers, self-commenting.

Up-to-date library documentation retrieval using Context7 MCP tools. Process THINK → RESOLVE → FETCH → APPLY. Use when fetching library docs, resolving package names to IDs, getting implementation guides, exploring API references. Provides package resolution strategy, trust score evaluation, token scaling (3K-20K), topic selection patterns.

Comprehensive feature artifact analysis for loading and understanding complete feature context. Use when needing to: (1) Load all feature documentation before implementation, (2) Build mental model of feature requirements and architecture, (3) Prepare context for downstream operations like checklist generation or task execution, (4) Analyze feature completeness and artifact availability. Triggers on: analyze feature, load feature context, prepare feature, check feature prerequisites, understand feature scope.