Skills.

Adversarial code review of a diff or PR to surface real bugs the author would actually want to fix, with a high bar and no theater.

AI is telling people about your product right now, and some of it is wrong. This finds out what it says and why it says it. Then it hands you a list of fixes, in order.

Code ships faster than anyone can read it now, and the bugs that slip past review sit quietly in production until a customer trips over them. This skill sweeps the whole codebase on your behalf, ignores the noise, and hands you the short list of real, proven bugs worth an engineer's attention. Can be automatically run on a schedule.

Feature development pipeline - research, plan, track, and implement major features.

Force a skeptical second pass on your own work. Because 'it should work' has never once been true.

Build a real-time support chat system with a floating widget for users and an admin dashboard for support staff.

Products get confusing in a specific way: the person who built it knows which option to pick and what happens after the click, and none of that knowledge ever made it into the interface. New users hit a vague button, a cleverly named feature, or a choice with no stated tradeoff, and they guess or quietly leave instead of asking. This finds those spots and proposes the smallest wording change that removes the guess, without burying the UI in helper text.

Make any repository work cleanly with Claude Code Desktop's preview + parallel git worktrees, regardless of language or framework.

AI builds a page and never opens it, so it ships a card with no visible edge, a header band with a hard seam nobody meant to draw, and text sitting just under the contrast line. This skill looks at what actually rendered and catches what you would catch yourself if you had the time to squint at every state and breakpoint. You get back a numbered fix list, ranked by severity, with a screenshot and an exact change for every finding.

Audit user-facing error messages in any codebase against an error-writing rubric. Walks through findings one at a time and proposes rewrites. Use this whenever the user mentions auditing error messages, reviewing error copy, fixing "Something went wrong" messages, error UX, error wording, error tone, or anything about how errors are worded — even if they don't say the word "audit."

Generate favicons from a source image

Generate a branded social media image announcing a new feature or update. Auto-detects brand from your codebase and captures via Playwright.

Final pre-commit sweep — review the session and code changes for anything worth codifying in CLAUDE.md or README.md.

Audits your codebase for the AI/LLM cost bugs that cause surprise bills (runaway retries, missing spend caps, uncapped output, agent loops that never stop) and gives you the exact fix for each.

Run in any codebase and your AI coding agent wires up per-page Open Graph images with ogkit.dev, using your site's own fonts, colors and logo, in whatever framework you're on.

Generate absurdly thorough project documentation covering local setup, architecture, and deployment.

Turns a finished worktree into three plain sentences about what it actually does for customers, plus the exact env vars, migrations, and setup you need to handle before you deploy it.

Deep research before planning. Launches parallel agents across docs, web, and codebase, then synthesizes findings into actionable context.

Audit a codebase against table-stakes security hygiene and verify the product's security/privacy promises actually hold up in code.

Most SEO tools tell you what to do. /seo does it: each run measures your site, ranks every possible action on one scale (fix a wrong claim, repair a broken redirect, refresh a page sitting at position 8, consolidate duplicates, write the next piece, build a free tool, get cited by ChatGPT), and executes the single best one. It scores its own past actions against your site's real numbers, so a month in, it knows which moves work on your site and which don't.

A self-contained SEO content engine that decides what you should publish next, researches and fact-checks it, then ships one publishable piece in your brand voice. Built for 2026 search: every piece carries genuine information gain and is optimized to rank in Google and get cited by AI answer engines like ChatGPT and Perplexity.

A full SEO sprint as a Claude skill. Pulls real keyword data from Ahrefs, drafts a roadmap doc tailored to your site, and then executes it phase by phase — shipping alternative pages, comparison pages, use-case pages, and pillar playbooks with internal-linking and schema checks enforced before each phase ships.

Your branch works, but it's carrying the debris of getting there: debug probes, half-wired surfaces, AI-flavored copy, and whatever you quietly broke on the way. Ship-check runs six passes in order, each as an isolated subagent so your context stays clean, landing the safe fixes automatically and holding every judgment call for one confirmation at the end. You get one report, one review, and a branch that's actually ready to open.

Whimsy finds the spots where your product feels sterile and hands you a pick-list of small, on-brand tweaks to its copy and interactions that make people smile without trying too hard.