tylergibbs1

Strip the tells that make copy read as AI-generated or look cheap. Use when writing or editing any prose, UI copy, headings, commit or PR text, or video text, and before shipping copy review it against this list.

Sequence new product or feature work as a headless "prove" phase that validates the core value on real inputs before any UI, microservices, or scaling. Use at the start of a new product or feature, or once a core engine exists and the user is about to add UI or infrastructure.

Creates precise, modern AWS architecture diagrams as PNGs or Mermaid, grounded in source code, infrastructure code, and current AWS guidance. Use when a user asks for an AWS architecture diagram, cloud topology, runtime or deployment view, failure or recovery flow, AWS service visualization, implementation handoff graphic, or a clean Figma/Linear-style 16:9 diagram with AWS icons.

Make every loading state structurally match its own route's real content so there is zero perceived layout shift. Use when building or reviewing any loading.tsx, skeleton, Suspense fallback, or first-sync/import progress UI. Covers per-route fidelity, accessibility, and never showing "done" or stale data while work is still running.

Decide what to build from real pain and market research, not guesses. Use when the user asks for feature ideas, "what to build next", "wow factor", latent demand, a roadmap, or a competitive gap analysis. Covers researching real pain, the painkiller-not-vitamin test, competitor gap-mapping, and filing the prioritized results as tracked issues.

Applies Linear's UI/UX and product philosophy (design as product judgment, not decoration) when designing, building, or reviewing interfaces and making product decisions. Use when framing a product problem, planning a redesign, deciding scope, reducing visual noise, designing AI/agent interfaces, debating customization vs opinionated defaults, or setting up product process (handoffs, OKRs, A/B testing, design reviews). Triggers on "Linear style," "calm interface," "opinionated software," "is this the right problem," "should we add a setting," "feature request vs need," "redesign strategy," "should this be reversible/undo," "is chat the right AI interface," "novelty vs familiarity." Peer-aware: where Linear's opinions have limits (platforms, APIs as contracts, metrics vs taste, keyboard vs discoverability) it names the boundary and which peer answer applies. Not for cloning the Linear visual look (dark mode, gradients, minimalist landing pages); this is about product method, not the look.

Orchestrate a multi-agent implementation: decompose a task, fan out subagents to implement each piece in parallel, and brief each one with the context it needs plus a mandate to pull fresh docs via Context7. Use when the user wants to "use subagents", "fan out", "parallelize implementation", "orchestrate agents", spin up workers, or build a feature that splits into independent workstreams. Covers Claude Code subagents, Claude Code dynamic workflows, and Codex subagents.

Push the current changes to a pull request, then launch a dynamic workflow that reviews the PR diff, adversarially verifies each finding, and fixes the confirmed ones, pushing the fixes back to the PR. Use when the user wants to "push to a PR and review", "open a PR then review and fix findings", "ship and review", or run a multi-agent review-and-fix pass over a branch. Pairs with orchestrating-subagents (which implements the change first).

Implementation details for accessible, performant interfaces. Use alongside vercel-design-guidelines for detailed technical guidance on interactions, animations, and forms.

Deploy applications and websites to Vercel. Use this skill when the user requests deployment actions such as "Deploy my app", "Deploy this to production", "Create a preview deployment", "Deploy and give me the link", or "Push this live". No authentication required - returns preview URL and claimable deployment link.

Enforce strict type safety, accessibility standards, and code quality for TypeScript/JavaScript projects using Biome's formatter and linter. Use when writing, reviewing, or fixing TS/JS/TSX/JSX code, checking for a11y issues, linting errors, type safety problems, or when the user mentions code quality, formatting, accessibility, or best practices.

Enforce modern React 19 and Next.js App Router patterns - server-first data fetching, minimal useEffect, Server Components, Server Actions, and form hooks. Use when reviewing React/Next.js code, migrating legacy patterns, or building new features with App Router.

Designs agentic AI systems by selecting the right architecture pattern and context engineering strategy. Triggers on "build an agent," "agent architecture," "agent design," "multi-agent," "agent pattern," "orchestration," "context engineering," "sub-agents," "agent workflow," "Ralph loop," "ReAct," "swarm," "coordinator pattern," "how should I structure my agent," "my agent loses context," "agent keeps failing on long tasks." For tool-level design, see designing-agent-tools.

Designs and builds CLIs optimized for AI agent consumption — machine-readable output, schema introspection, input hardening, context window discipline, and safety rails. Triggers on "CLI for agents," "agent CLI," "agent-first CLI," "make my CLI agent-friendly," "MCP server from CLI," "machine-readable output," "agent DX," "CLI tool for AI," "rewrite CLI for agents," "agents keep breaking my CLI," "input validation for agents." For tool schema design, see designing-agent-tools. For prompting, see prompting-agents.

Designs, builds, and iteratively improves tools for AI agents — covering schema design, descriptions, response formatting, MCP servers, programmatic tool calling (PTC), progressive disclosure, evaluation, and evolving tools with model capabilities. Triggers on "build a tool," "write a tool," "MCP tool," "agent tool," "tool for Claude," "tool design," "improve my tools," "tool evaluation," "tool description," "programmatic tool calling," "PTC," "tool composition," "reduce tool call tokens," "progressive disclosure," "too many tools," "my agent keeps calling the wrong tool," "agent can't use this tool," building MCP servers, Desktop extensions (DXT), or API tool definitions.

Writes and optimizes prompts for AI agents and agentic systems — system prompts, tool descriptions, thinking configuration, output formatting, and multi-window state management. Principles are model-agnostic; examples use current Claude models. Triggers on "write a system prompt," "agent prompt," "prompt engineering," "my agent keeps doing X wrong," "how do I prompt," "system prompt," "tool prompt," "thinking tokens," "effort setting," "agent is too verbose," "won't use tools," "overthinks," "reduce hallucinations," "prompt for agent," "agentic prompt." For tool schema design, see designing-agent-tools. For agent architecture, see designing-agents.

Leave notes on files for other agents to see in future sessions. Use after making non-obvious changes, fixing tricky bugs, or when code looks wrong but is intentional.

Leave notes on files for other agents to see in future sessions. Use after making non-obvious changes, fixing tricky bugs, or when code looks wrong but is intentional.

Orchestrate a multi-agent implementation: decompose a task, fan out subagents to implement each piece in parallel, and brief each one with the context it needs plus a mandate to pull fresh docs via Context7. Use when the user wants to "use subagents", "fan out", "parallelize implementation", "orchestrate agents", spin up workers, or build a feature that splits into independent workstreams. Covers Claude Code subagents, Claude Code dynamic workflows, and Codex subagents.

Push the current changes to a pull request, then launch a dynamic workflow that reviews the PR diff, adversarially verifies each finding, and fixes the confirmed ones — pushing the fixes back to the PR. Use when the user wants to "push to a PR and review", "open a PR then review and fix findings", "ship and review", or run a multi-agent review-and-fix pass over a branch. Pairs with orchestrating-subagents (which implements the change first).

Runs an autonomous test generation loop that finds uncovered code paths, writes tests for them, runs the suite, and keeps tests that pass and increase coverage. Loops indefinitely. Triggers on "auto-test", "generate tests", "increase coverage", "write tests for my codebase", "test generation loop", or "cover untested code".

Generates edge-case inputs for functions, runs them, catches crashes and unexpected outputs, applies minimal fixes, and adds regression tests. Loops until no new failures are found. Triggers on "fuzz", "fuzz-fix", "harden my code", "find edge cases", "crash test", "find bugs with fuzzing", or "harden my API". Good for hardening APIs, parsers, validators, and any function that handles untrusted input.

Use when the user asks you to build, edit, redesign, or export an AI-generated slide deck in Substrate — the desktop app with a local MCP server at http://127.0.0.1:4321/mcp. Triggers on: any substrate:* tool being available, "build a deck in Substrate", "make a slide deck", a deck/slide the user is editing in Substrate, or "open Substrate". Covers the build loop, writing render-quality slide prompts, applying brand designs, the propose/review flow, and known gotchas. Out of scope: editing pixels — Substrate has no pixel/layer surface by design.

Runs an autonomous experiment loop that modifies code, measures a target metric, keeps improvements, and discards failures. Operates indefinitely without human intervention. Triggers on "auto research", "run experiments", "optimize autonomously", "research loop", or "find improvements automatically".

Applies Evil Rabbit's (Vercel's founding brand designer) and the Geist design system's approach when designing, building, or reviewing interfaces, brand, and typography. Use when designing UI, building a marketing or product surface, choosing color/type/spacing, reviewing visual craft, designing a logo or brand mark, deciding whether a flourish earns its place, or matching the Vercel/Geist aesthetic. Triggers on "Vercel style," "Geist," "make it look like Vercel," "monochrome design," "minimal developer-tool aesthetic," "Swiss/grotesque type," "hairline borders," "design system tokens," "is this decorative or functional," "brand mark," "high-contrast minimal UI."

Get fresh, live LLM pricing and cost estimates via the tokenomics CLI or MCP server.

Runs an autonomous ML research loop on a remote RunPod GPU over SSH that forms hypotheses, runs experiments, measures multiple capability probes, and hunts for emergent properties — capabilities that appear unexpectedly rather than being explicitly trained. Keeps reproduced discoveries, discards noise, and logs everything to a research journal. Use when the user wants to find emergent behavior, run open-ended ML experiments, search for novel or revolutionary capabilities, or mentions "find emergent properties", "emergence research", "autonomous ML experiments", or provides a RunPod for experimentation.

Turn a single concept into a TikTok-style "caveman explains it" slide deck — one idea per slide, 3 to 5 words of big bold text, clean flat illustrations on a pure white background, built with the Substrate MCP. Use when the user wants to make a TikTok/Reels/Shorts explainer, a "caveman voice" or "explain like I'm dumb" deck, a punchy slide-per-beat video script, or says "make a deck about X" for short-form video. The creator narrates in caveman voice over the slides, so slides carry minimal text and one idea each.