parvez

Score whether a behavioral design pattern (Strategy, State, Command, Observer, Chain of Responsibility, Template Method, Visitor, Mediator, Iterator, Memento) fits a code site, and recommend it only when scored preconditions are met. Use when the user asks "should I use a strategy/state/observer here", "where can I apply a behavioral pattern", or when object interaction is awkward (switch on a type to pick an algorithm, behavior driven by internal state, requests needing queue/undo, one-to-many notifications, request handled by a chain, duplicated algorithm skeletons, operations spread across a class hierarchy, tangled many-to-many communication). Each pattern has opportunity signatures, weighted preconditions, a threshold, and hard-veto anti-indicators; emits APPLY / CONSIDER / REJECT with cited evidence. Spoke of refactoring-guru; usable standalone.

Use when asked to test a branch, pull request or ticket against the real running app and hand back proof - "run the tests and give me a report", "test this branch end to end", "automated test with visual verification", "QA this with screenshots", "give me a test report for this PR". Runs the backend tests, seeds and cleans test data, drives the API and the browser, and writes one self-contained HTML report with screenshots that a product owner can read. Produces a run against the live app, not a reading of the diff.

Drive ONE reported bug from report to verified fix, with the root cause proven before any fix is written. Use when the user reports a bug, defect, regression, wrong or missing data, a 500, a stack trace, "it broke", "it was working yesterday", or hands over a bug ticket and wants it taken end-to-end. Triggers on "/bug-fixer", "fix this bug", "why is this broken", "root cause this", "this wrote wrong data". One command, one bug, root cause before fix.

A rulebook + bootstrap + guard for building software with Clean / Hexagonal (Ports-and-Adapters) Architecture, in any language (.NET, TypeScript/Node, Java/Kotlin, Go, Python). Use this when starting a new project and you want to lay out the layers correctly, when structuring or refactoring code into Domain / Application / Infrastructure / Presentation, when deciding "where does this code go?", when adding a dependency or project reference and unsure if it inverts the architecture, or when the user asks "set up clean architecture", "how should I structure this", "is this the right layer", "did I break the dependency rule", "bootstrap a hexagonal project". Ships a config-driven layering guard so the boundaries are enforced, not just hoped for.

Detect code smells in a file, function, or directory and map each to candidate fixes. Use when the user wants to find what is wrong with code, "find code smells", "review this for quality", "what should I clean up here", or as the first step of a vague "refactor this" / "improve this" request. Covers the 22 refactoring.guru smells across five groups (Bloaters, Object-Orientation Abusers, Change Preventers, Dispensables, Couplers). Reports each smell with cited file:line evidence and the candidate refactoring or design pattern it maps to. Spoke of refactoring-guru; usable standalone.

Score whether a creational design pattern (Factory Method, Abstract Factory, Builder, Prototype, Singleton) fits a code site, and recommend it only when scored preconditions are met. Use when the user asks "should I use a factory/builder here", "where can I apply a creational pattern", "how should I construct these objects", or when object construction is awkward (telescoping constructors, scattered `new`, type-switch instantiation, expensive clones, hand-rolled globals). Each pattern has opportunity signatures, weighted preconditions, a threshold, and hard-veto anti-indicators; emits APPLY / CONSIDER / REJECT with cited evidence. Spoke of refactoring-guru; usable standalone.

Build a pixel-exact, clickable HTML demo of a planned workflow that looks exactly like the app's real components - web via a DOM + CSS snapshot of the running, signed-in app; desktop (WinForms) via the app's own controls rendered off-screen. Use whenever the user asks for an HTML demo, clickable mockup, prototype, "pixel perfect demo", "how exactly will this workflow look", "show the PO / stakeholders how it will look", or a demo for a ticket or user story in a web app, a desktop client or both - even if they do not say "skill". Not for free-form design exploration or Figma work.

Stand up a "grill room": independent, code-grounded agents that interrogate shipped features at the PROMISE level (does the real user-facing promise actually hold in the code, not "does it compile"), plus an optional CEO cost evaluator for unit economics. Each interrogator asks fundamental questions and ANSWERS them by reading the real code, then returns a verdict (BULLETPROOF / CRACKED / FAKE) with file:line evidence. Use whenever the request is to "grill" or pressure-test a feature, "is X actually ready / bulletproof", "can users really do Y" (e.g. publish to a third-party platform, download the export, does the stitch seam hold), audit promise-vs-reality across a codebase, interview features by reviewing code, make features bulletproof, or evaluate whether the cost/margin is too much. One feature (inline) or a whole release/phase (workflow sweep).

Design and verify service health endpoints that report REAL readiness (database and critical dependencies reachable) instead of a hardcoded 200, and wire them correctly into deploy gates, container HEALTHCHECKs, and orchestrator probes. Use this whenever you add or review a /health or /healthz endpoint, set up a Docker/Kubernetes/compose health check or a post-deploy probe, or when the user asks "why did the deploy go green but the app is down?", "add a readiness probe", "make the health check honest", or is debugging a service that's "up" but not actually serving. A health endpoint that returns 200 unconditionally silently defeats every gate built on it - so make it assert something true.

House style for any prose that lands in git history or on a reviewer's screen - PR descriptions, commit subjects and bodies, PR comments, code comments, README and docs prose, ticket notes. Enforces ASCII punctuation only, Simplified Technical English (ASD-STE100), no filler, no jargon, and comments that only explain a non-obvious why. Carries a PR-description sub-skill that finds and fills the repo's own PR template section by section. Use BEFORE writing or editing such prose. Triggers on "write the PR description", "fill the PR template", "draft the PR body", "write a commit message", "write this up", "clean up this comment", "make this readable", and on any request to generate text destined for a commit, a PR, a code comment or repo docs.

Orchestrator for code refactoring and design-pattern work. Use when the user wants to improve, clean up, restructure, or "refactor" code; find code smells; decide whether or where to apply a design pattern (Factory, Builder, Strategy, Observer, Adapter, etc.); or asks "is this pattern right here", "where can I use a pattern", "how do I fix this smell". Understands the user's intent, then routes to the right spoke skill (code-smell-detector, refactoring-methods, creational-patterns, structural-patterns, behavioral-patterns) and merges their findings. Patterns are only recommended when a code site meets scored preconditions with cited evidence.

Apply Martin Fowler's refactoring techniques to fix a known code smell. Use when the user knows what is wrong and wants the safe step-by-step mechanics to fix it, "how do I extract this method", "how do I break up this class", "replace this conditional with polymorphism", "introduce a parameter object". Holds the refactoring catalog grouped as composing methods, moving features, organizing data, simplifying conditionals, and simplifying method calls, each with preconditions, ordered mechanics, and a safety rule. Recommends and gives mechanics only; never auto-edits. Spoke of refactoring-guru; usable standalone.

Audit and improve a website's SEO (search engines) and AEO (answer engines - ChatGPT, Perplexity, Google AI Overviews, Copilot, voice assistants), then prove the result with local tests. Use this whenever the user wants a site, landing page or marketing page to rank better, be found or cited by AI assistants, show correct previews when shared, or asks about meta tags, JSON-LD / schema.org, robots.txt, sitemap.xml, llms.txt, canonical URLs, noindex, prerendering or SSR for crawlers, Core Web Vitals / Lighthouse for a public page, featured snippets, FAQ schema, or "does our site follow this SEO/AEO article". Also use it when the user pastes an SEO or AEO blog post and asks whether the site does what it says. Works on any stack (React/Vite SPA, Next.js, static HTML, CMS) and on a live URL or a local build.

Use when someone wants to see how much time, how many sessions, or how many tokens and dollars they have spent on a coding project locally. Triggers on "show usage", "how long have I worked on this", "how much time did I spend", "my claude time", "wakatime for claude", "session stats", "token usage", "cost per project", "how many hours on this project". Reads local Claude Code transcripts and prints WakaTime-style active time plus token/cost/tool tables. Local, read-only, zero deps.

Drive ONE user story, feature request or ticket from raw ask to verified, review-ready code: skeptical requirement analysis, code exploration, task breakdown, user picks tasks, implement, test, verify. Use when the user hands over a single scoped feature and wants the whole process driven, not a one-off edit. Triggers on "/story-implementer", "build this story", "implement this ticket", "drive this feature", or a single scoped feature ask. One command, one user story.

Score whether a structural design pattern (Adapter, Bridge, Composite, Decorator, Facade, Flyweight, Proxy) fits a code site, and recommend it only when scored preconditions are met. Use when the user asks "should I use an adapter/decorator/facade here", "where can I apply a structural pattern", or when object composition is awkward (incompatible interfaces, subclass explosion across two dimensions, part-whole trees, runtime-added responsibilities, message chains into a subsystem, memory blowup from many similar objects, controlled access). Each pattern has opportunity signatures, weighted preconditions, a threshold, and hard-veto anti-indicators; emits APPLY / CONSIDER / REJECT with cited evidence. Spoke of refactoring-guru; usable standalone.

Use when the user needs to fill in a timesheet or time log and asks what they worked on for one or more past dates - "I need to fill my timesheet for these days", "what did I do on the 13th and 14th", "fill up my time log for last week", "search my conversations and check what I did on Monday". Reconstructs ticket numbers, short descriptions and non-overlapping time blocks per day from local Claude Code session transcripts under ~/.claude/projects, ready to paste into a time tracker.

Write a user-facing feature guide for a ticket, issue or pull request as one self-contained HTML file in a fixed, print-ready format, check every claim in it against the code with a fresh agent, print it to PDF, and attach it to the tracker item (GitHub, Jira, Azure DevOps or other). Use whenever the user asks for a "user guide", "feature guide", "guide for this ticket / card / PR", "end-user docs for this feature", wants such a guide as HTML or PDF, wants it attached to the ticket, or asks to update, re-verify or re-attach a guide after the code changed - even if they do not say "skill".

Make animated promo, explainer, product-launch, feature-announcement, tutorial-teaser or social-media videos entirely from code, as a ready-to-publish MP4 with voice-over, music, captions and a thumbnail. Use this skill whenever the user asks for a video, promo, teaser, ad, explainer, reel, short, TikTok, LinkedIn or YouTube clip, motion graphic, animated walkthrough, or "a video about this project/product/feature" - even if they do not name any tool - and whenever they want to change, lengthen, shorten, re-colour, re-voice or re-render a video made this way.