Production-Tested Claude Skills

7 skills that speak JSON to your backend.

Product SEO, programmatic collection pages with a hard thin-content threshold, scheduled blog posts that never repeat a topic, drift audits against previous runs, and a video-to-comparison-article chain. Multi-language, with reciprocal hreflang built in.

Eleven modules, one real launch order.

From empty repo to analytics: security from day one, a VPS deploy that ships only changed files, SEO foundations as pages are written. Every command was executed on a production launch, including the mistakes and their fixes.

Eleven skills that audit the whole stack before you ship.

Database, performance, data protection, secure logging, accessibility, test strategy, threat modeling, API design, and legal content. One orchestrator skill runs them all in two passes and produces a single matrix report. The breadth companion to the security-audit pack, which goes deep on application security.

Behavioral emails, image coaching, and an advisor that finds what you forgot.

email-drip covers 11 scenarios, each with its own trigger, timing, and exit condition. product-image-analyzer coaches sellers during upload without blocking. product-advisor reviews your platform like an owner.

Eight skills for the daily loop.

Reviews that report real defects with concrete failure scenarios, commit messages derived from the actual diff, verified READMEs, changelogs, PR descriptions, an environment doctor that applies the smallest fix, plus element-registry, which hunts down duplicated components and keeps a component-guide discipline. The eighth is replacement-auditor: it makes you prove the old implementation actually died when the new one landed.

8 layers, ~30 items, adapted to your real stack.

Ten injection variants in one table, business logic, sessions, monitoring, infrastructure, backups, legal compliance, and supply chain. Every item ends in a verdict: in place, fixed, or N/A with a reason.

Free-text idea in, design doc and risk-scaled prompts out.

Breaks a product idea into workflows, roles, and open questions, then produces a Claude Code prompt set where payment and auth modules get smaller, test-gated steps.

Production-grade payments for the Turkish market.

The complete PayTR iFrame flow: server-side hash verification with constant-time compare, idempotent callbacks, an append-only raw-notification journal, amount verification that closes the repriced-cart attack, and a go-live checklist. Ships with 8 failure modes seen in production and their fixes.

The architecture data set a marketplace needs before the build starts.

Actor × capability matrix, an entity model with the ownership and publication columns spelled out, the interaction model for ratings, messaging and reputation, and where each rule is enforced. Ships with the failure catalogue: split reputation, thread keys without a subject, hard-coded option lists, permission gates that live only in the UI.

Four gates that stop an agent-built project from rotting.

The new thing must kill the old one, your proof must actually prove, schema and code and deploy have a fixed order, and the agent knows what it may do unasked. Behind them sits a 12-case failure catalogue: the flag nothing writes, the policy that loosened access instead of closing it, the deploy that kept dead files alive, the benchmark that measured its own escape branch. How many would a test suite catch?

An audit that measures how AI-generated your interface looks.

The indigo gradient, Inter, the three-card grid: a model's statistical defaults make every interface interchangeable. This skill lints the code, scores it, and returns line-level fixes. It does for design what the free humanize-writing does for prose.

The anti-slop skill, free.

Banned openers, banned hedges, and translation contamination checks with bad/good pairs, plus a generate→validate→retry→skip loop for automated pipelines.