Claude Skills by khasky

Audits UI and markup for accessibility (WCAG 2.1/2.2, keyboard, screen readers) and suggests concrete fixes. Use when checking a11y, before shipping a page or component, reviewing forms/modals/interactive UI, or when the user says 'accessibility', 'a11y', 'WCAG', 'screen reader', 'keyboard navigation', 'доступность'. Covers semantic HTML, focus management, labels, contrast, dynamic content, and WCAG 2.2 additions (target size, focus obscured, accessible authentication). Do not use for convers...

Designs or reviews the shape of an HTTP API before code exists — resource modeling, versioning strategy, pagination, idempotency, filtering, and how the error contract plugs in. Use when asked to 'design the API', 'review my API design', 'how should we version / paginate', 'add an endpoint' at the design stage, or when awesome-design-doc needs the API contract detailed. Do not use for the error envelope internals and retry semantics — use awesome-error-standards; not for auth/injection review...

Read-only whole-project deep audit — architecture and module boundaries, documentation-vs-code fidelity, design-principle adherence (YAGNI/KISS/SOLID), and extensibility (how easily someone builds from source and adds a new provider/plugin/adapter) — producing a prioritized report with a SHIP / FIX / BLOCK verdict. Framework- and language-agnostic. Use when the user asks to 'audit/analyze the architecture', 'review the whole codebase design', 'does the code match the README/docs', 'how hard i...

Debugs by building a runnable pass/fail reproduction, isolating the root cause, then fixing — no fixes without root cause first. Use when fixing bugs, investigating test/build failures, or when the user reports an error, crash, flaky behavior, or says 'debug this', 'why does X happen', 'не работает'. Use especially under time pressure or after several failed fix attempts. Do not use for feature work with no failing behavior to explain.

Audit every checkable public claim a product makes against the code that decides it, then fix the drift. Compares marketing pages, README, store listings, docs, privacy policy, FAQ and structured data against the constants, manifests, catalogs and locale strings that settle them — including across repos, where nothing fails when a sentence goes false. Use when asked to 'check the site against the code', 'do the facts on the site match reality', 'audit the copy', 'find outdated claims', 'is th...

Repo-wide cleanup of AI-like code noise: comments to a strict bar (delete narrating, condense bloated, fix stale, keep load-bearing), a read-only audit mode, an opt-in refactor mode for vague names, dead or duplicated code and over-abstraction, and a detection mode for dead code and logical inconsistencies with execution-verified findings; cleanup stays behavior-preserving. Use when asked to clean up, prune or condense comments, remove "obvious" or AI-generated comments, make code self-docume...

How to respond to code review feedback: verify before implementing, clarify unclear items, and push back with technical reasoning when needed. Use when receiving PR feedback or review comments, when reviewer or bot findings need addressing, or when the user says 'address the review', 'ответь на ревью', 'поправь по замечаниям'. Applies to any team or culture. Do not use for performing the review itself — use awesome-code-review.

Reviews diffs and PRs for correctness, security, and team standards, with confidence-scored findings in severity buckets (Critical / Suggestions / Nice to have). Use when reviewing a pull request, merge request, patch, or diff; before merging; after completing a feature; or when the user says 'review this PR', 'check this change', 'review my diff', 'сделай ревью'. Do not use for responding to review feedback you received (use awesome-code-review-feedback) or for docs-only and formatting-only ...

Universal coding standards: naming, structure, immutability, error handling, type safety, plus backend layering and frontend architecture/motion patterns for consistent code. Use when starting a project or module, refactoring to team conventions, setting up lint/format rules, onboarding, or when the user says 'coding standards', 'naming conventions', 'code style', 'стандарты кода'. Discovers and follows the project's own conventions first; these rules fill the gaps. Do not use for a repo-wide...

Turns a codebase into a commit plan: a navigation map of its modules and their dependency direction, then a split where every commit builds and tests on its own, so the series is bisectable end to end. Each commit is verified by replaying the ladder in a scratch clone against the repository's own gates, never asserted. Messages follow a strict ruleset that keeps machine-written prose out of the log. The only output is a plan file numbered #1 to #N, no dates, no preamble. Use when asked to 'sp...

Builds a scheduled batch of platform-fit content-marketing posts from any sources (repositories, websites, files, code): a knowledge map with every claim traced to evidence, an interview for voice, frequency, duration, platforms and timezone, dated checks of each platform's limits, posts written to a per-genre register and an author voice profile when one exists, offline HTML/CSS graphics when a platform needs media and has none, a two-stage self-audit against AI-slop tells, one dated file pe...

Publishes a prepared batch of scheduled posts to the user's own accounts through their live browser (Playwright MCP --extension bridge to Chrome or Edge): a preflight that names which browser and profile it attached to, post-source scan with hard-stop format validation, per-platform login checks with a wait-or-skip choice (login is never automated), a persistent ledger that survives restarts and prevents duplicate posts, timezone-mapped scheduling that can idle for days, strictly sequential h...

Repurposes one existing text — a published URL, a file, or pasted notes — into platform-native posts, then files them and optionally publishes: source notes every claim traces back to, an interview for platforms, voice, language and length, per-genre registers and dated platform limits reused from awesome-content-campaign, a two-stage anti-slop audit, and one dated file per platform in the format awesome-content-publisher takes, so publishing is a handoff, not a second implementation. Use whe...

Builds a durable, reusable author-voice profile from whatever evidence exists — the user's own posts read through their logged-in browser, files of past writing, pasted samples, a website (via awesome-style-mimic), or an interview alone — and writes one profile file other skills read instead of re-deriving a voice per task: presence signals, absence signals with per-sample counts, per-platform register, protected personal tics, verbatim samples, and a source-and-confidence stamp. Thin evidenc...

Read-only audit of a database layer — schema design against a catalog of SQL anti-patterns (EAV, generic keys, imprecise types), query patterns (SELECT *, N+1, unindexable predicates), integrity and concurrency (constraints in the DB, transactions, locking), and migration/operations hygiene (forward-only, expand/contract, restore path, tenancy model) — producing evidence-backed findings and a SHIP / FIX / BLOCK verdict. Use when the user asks to 'audit the database', 'review the schema', 'che...

Read-only audit of a project's third-party dependencies — lockfile discipline, hallucinated and typosquatted package names (slopsquatting), dependency confusion, install-script exposure, maintenance and provenance, license conflicts, and CVE reachability — producing evidence-backed findings and a SHIP / FIX / BLOCK verdict. Use when the user asks to 'audit dependencies', 'check the supply chain', 'is this package safe', 'проверь зависимости', to review a package.json/requirements/go.mod chang...

Plans and executes dependency upgrades safely — batching by risk, changelog-driven major migrations, overrides for unfixed CVEs, verification between steps, one revertable commit per batch. Use when asked to 'upgrade dependencies', 'bump packages', 'fix the audit findings', 'update to the latest version', 'обнови зависимости', or to act on an awesome-dependency-audit report. Do not use for detection and risk assessment — use awesome-dependency-audit; adding a brand-new dependency is an ask-fi...

Produces a design document or ADR for a feature or architectural decision: requirements and scale numbers first, real alternatives with trade-offs, a recommendation grounded in the requirements, explicit non-goals and migration path. Use when asked to 'write a design doc', 'write an ADR', 'design this feature/system', 'which approach should we take', 'спроектируй', or when awesome-code-review requests an ADR for a load-bearing decision. Do not use for auditing an existing architecture — use a...

Clean and line-edit Markdown documents for clear, specific, publication-ready prose. Use when asked to de-bloat, de-template, remove chatbot artifacts, fix vague wording, settle whether numbers are written as digits or spelled out, or make Markdown read like careful human editorial writing. Do not use to fabricate facts or bypass detectors.

Applies consistent error handling, logging, and user-facing messages: typed errors, operational-vs-programmer classification, API error envelopes, HTTP status mapping. Use when adding or refactoring error handling, designing an API error contract, reviewing failure paths, or when the user says 'error handling', 'consistent errors', 'обработка ошибок'. Do not use for log format, levels, or PII redaction — use awesome-logging-standards.

Rewrites the author and committer identity on a commit — or on every commit carrying a wrong identity — and force-pushes it, safely: account-ownership and write-access checks, a verified mirror backup, a counted hash blast radius, and an explicit confirmation gate before anything irreversible. Takes a commit URL plus the replacement identity: '/awesome-git-author-rewrite <commit-url> --name '<Name>' --email '<email>''. Use when asked to 'fix the commit author', 'wrong email on my commits', 'c...

Erases a repository's git history and rebuilds it as a curated commit series over the same tree: a split plan the user approves (proposed here, or supplied by awesome-commit-plan and validated first), the repo's own commit rules obeyed (commitlint, hooks, CONTRIBUTING), paced timestamps anchorable to the earliest evidenced activity (recovered via awesome-git-history-salvage when refs fall short), signed commits, and safety throughout: a verified mirror backup, a confirmation gate before every...

Erases a repository's entire git history and replaces it with a single fresh commit, then force-pushes — safely: access checks, a verified mirror backup, a secret scan, and an explicit confirmation gate before anything irreversible. Use when asked to 'wipe git history', 'squash all commits into one', 'reset history to a single Initial commit', 'start the history fresh', 'clean/erase the commit log', or in Russian 'очистить историю git', 'схлопнуть все коммиты в один', 'переписать историю с ну...

Reconstructs every commit a repository has ever held, including history a force-push erased, by merging current refs, pull-request refs, any mirror backup and every ref state the host's activity log recorded, then fetching unreachable commits by SHA over the git protocol. Emits sha/date/author/subject rows flagged for survival on the default branch, at the detail level the user picks: per source, per sha, or per logical commit with rewrite twins collapsed. Use when asked 'what was in this rep...

Advisory copy-edit of prose — reports grammar, logic, and flow issues as suggestions without rewriting the text. Use when the user asks to 'check grammar', 'proofread', 'copy-edit', 'find errors in', or 'review the writing' of an article, email, doc, or post, and wants to keep control of the wording. Not for removing AI-generated voice/markers (use awesome-humanize-en) and not for rewriting or de-bloating Markdown in place (use awesome-document-style).

Use when: the user asks you to check, rewrite, humanize, or de-slop English text that shows signs of AI generation (ChatGPT, Claude, Gemini, Grok, DeepSeek, GPT-5, Qwen, Llama, any LLM). Trigger for requests like: 'humanize this', 'make it sound human / natural', 'remove the AI voice', 'this reads like a chatbot', 'AI detector for English', 'check for AI', 'de-slop', 'reduce AI-isms', remove clichés / filler / corporate jargon / emoji bullet lists 🚀 / gratuitous em-dashes / sycophantic tone ...

Read-only, mechanical audit of a landing or marketing page's conversion structure — single primary CTA per view, form-field friction, ad-to-headline message match, trust/social-proof presence, above-the-fold value, and CLS-safe banners/interstitials — producing evidence-backed findings and a SHIP / FIX / BLOCK verdict per page. Use when the user asks to 'audit my landing page', 'why isn't this page converting', 'check the CTA / form / hero', or 'is this page conversion-ready'. Do not use for ...

Audit a public-facing client (browser extension, mobile/desktop app, SPA, CLI, SDK, or any open-sourced client that talks to a private backend) so its public surface stays self-contained and doesn't help attackers. Use when asked to review a repo for leaked backend internals, secrets, or abuse-enabling disclosure; to check that comments/docs/tests don't reveal server-side mechanics (rate limits, anti-abuse, quotas, test backdoors, infra/tech stack, DB/schema, env-var names); to scrub a client...

Applies structured logging, levels, PII handling, and wide-event (canonical log line) patterns. Use when adding or reviewing logs, choosing log levels, designing request logging, after an incident where logs were insufficient or leaked data, or when the user says 'logging', 'log format', 'what should we log', 'логирование'. Do not use for designing the error contract or retry policy itself — use awesome-error-standards.

Runs an authorization-gated penetration-test engagement against a target (repo, web app, API, network, cloud, or mobile client) following PTES, OWASP WSTG/MASTG/API Top 10, and NIST SP 800-115 — from scoping and rules of engagement through recon, attack-surface mapping, testing, exploitation-to-proof, and a findings report with CWE/CVSS and retest. Passive review of your own code needs no gate; any active probe, scan, or exploitation is hard-gated behind written authorization and scope. Use w...

Read-only audit of performance and reliability — event-loop discipline, streaming and backpressure, memory/CPU diagnostics, production shutdown/timeout/job habits, cross-service resilience topology (circuit breakers, retry budgets, queue bounds, dual-writes), and frontend delivery (Core Web Vitals, bundle size, hydration) — producing evidence-backed findings and a SHIP / FIX / BLOCK verdict. Use when the user asks to 'audit performance', 'why is the service slow', 'why is my page slow', 'memo...

Read-only regression sweep over a codebase and its live surfaces: a layered aspect pass (typecheck, lint, generated-artifact drift, unit and integration suites, black-box wire contract, cross-implementation parity, downstream consumer build) compared against a recorded baseline so the answer is deltas, plus nine rotating deep angles — shuffled test order, the other publication path, coverage ratchet, deploy dry-run, one number by two paths, docs-vs-code constants, deployed-vs-committed drift,...

Structured root-cause analysis for incidents, recurring failures, and process problems that have no failing test to run — 5-Whys, fishbone, PDCA, and a one-page A3 write-up. Use for a postmortem, incident retro, 'why does this keep happening', quality or process breakdown, or 'find the root cause' of a non-code issue. For an actual code bug with a reproducible failure, use awesome-bug-fix instead.

Audits code for common vulnerabilities: injection, secrets, auth, dependency CVEs, CI/CD pipeline exposure, and cryptographic misuse — with confidence-gated, evidence-backed findings mapped to CWE/OWASP. Use when reviewing security, before a release, after adding auth/payments/sensitive-data handling, when hardening GitHub Actions or other CI workflows, or when the user says 'security review', 'security audit', 'check for vulnerabilities', 'is this secure'. Do not use for auditing what a publ...

Read-only SEO and AI-discoverability audit of a site or codebase — technical SEO, generated-page (programmatic SEO) safety, and agent/LLM readability — producing evidence-backed findings and a SHIP / FIX / BLOCK verdict. Use when the user asks to 'audit my SEO', 'check for thin content / cannibalization', 'will Google penalize these pages', 'am I blocking GPTBot/ClaudeBot', 'llms.txt', 'AI discoverability', 'will LLMs see my site', 'is my site agent-readable', or 'проверь SEO'. It audits and ...

Removes installed agent skills from every AI agent on the machine — Claude Code, Codex, Gemini CLI, Cursor, Copilot, opencode, Amp, Windsurf, and any other agent that reads a `skills/` directory — behind a keep list that can spare one collection, this skill alone, or a named set. Cross-platform by construction: the platform is detected at run time and every command comes from that detection. Use when asked to 'delete all skills', 'uninstall every skill from all agents', 'clean out my agent sk...

Audits a repository for machine-written 'AI slop' markers across every surface — code, comments, tests, docs, configs and CI — against a verified catalog: glyph pockets, stale and false comments, change-narration, drift-bait numbers, template stamps, misleading prefixes, impossible defensiveness, copy-paste drift, em-dash saturation and negative-parallelism prose fingerprints. Every suspect is verified against the code before it is reported, absence is proven per category, and an opt-in fix p...

Learns a website's writing style by deep-crawling it in a live browser and distills a reusable style guide (voice, tone rules, lexicon, rhythm, structure, golden samples), then rewrites any documents or page sources in that voice with a cross-document consistency pass. Use when the user says 'learn the style of <site>', 'изучи стиль сайта', 'create a style guide from my website', or wants text rewritten in a learned voice: 'rewrite this in the style of <site>', 'перепиши в стиле', 'apply this...

Designs and writes tests that catch real regressions — placement (unit/integration/E2E), factories and fixtures, behavior-first assertions, characterization tests for legacy code, property/fuzz tests for parsers. Use when asked to 'write tests', 'add test coverage', 'test this module', 'напиши тесты', when a bug fix needs its regression test, or when awesome-bug-fix, awesome-code-review, or awesome-security-audit calls for a failing or regression test. Do not use for judging existing tests in...

Translate Russian text of any kind into natural English while preserving every structural element, formatting marker, link, identifier, and the author's human voice. Use when translating Russian articles, docs, prose, fiction, marketing copy, technical content, transcripts, or any other Russian-language text to English; when mirroring a directory of Russian files into an English equivalent; or when the user asks to \"translate to English\", \"convert to en\", \"make an English version\", \"пе...