storybook
claudecodexBuild, preview, and test UI components with Storybook.
7 skills
도메인 중립 광고 이미지 소재 프롬프트 생성 시스템 — 공개 광고 투명성 라이브러리(Meta Ad Library + Google Ads Transparency) 수집 기반, prompt-only. orchestrator + 24 활성 서브에이전트.
Plugin marketing-img from marketplace mwroh-dev in mwroh-dev/marketing-ad-img. Commands read from the repo's manifest — 1 month ago.
This plugin runs code on your machine — its hooks fire automatically on events like tool use and it starts local MCP servers. The exact commands are listed below; read them before installing.
agent-browser-explorationExplore an unknown browser source with agent-browser and produce a discovery report and candidate flow notes only.
user-answer-toolingConvert each raw interview answer into a structured user-answer artifact and update interview-state slot values.
/marketing-img:startGenerate ad-image creative prompts through the marketing-img orchestrator.
ad-analystExtracts and normalizes ad keywords from a per-persona competitor corpus (titles + detail text), labels each by functional ad-slot, and collapses transliterated Korean loanword variants into a single canonical. Outputs keyword instances only; ranking is a deterministic script. Use after competitor d
ad-creative-refinerClassifies ONE detail-cut image (the seller's own / user-provided) into a TYPE (ad_creative / catalog / spec / review / lifestyle / unknown) so the refiner gate can separate persuasion detail-cut ad creatives from non-ad images. STRICT: ad_creative ONLY when persuasion copy (headline/benefit) is com
ad-image-screenerad-intent-deriverDerives the TYPED ad-intent draft from the confirmed intake slots (plain seller words → intent_stage/objective[]/occasion[]/what/who + sourced claim_constraints). Use at request-evaluation step 4, when ready=true, BEFORE the recognition gate. Owns the derivation judgment the orchestrator must not se
ad-type-classifierClassifies ONE ad into its established advertising TYPE — message_basis (informational/transformational/hybrid, Puto & Wells 1984), execution_style (testimonial/demonstration/lifestyle/…, Belch & Belch / Kotler), and the routed ad_type adapter — so the analysis stage applies a per-type gate-check (d
brand-researcherFor initial-setup — researches ONE brand/product from PUBLIC sources (the user-provided product page, public reviews/comments, web search) and returns evidence-grounded category + persona candidates the interview presents as CHOICES. One research angle per dispatch (page | reviews | positioning) so
competitive-analystWrites the interpretive synthesis ON TOP OF the deterministic competitive-trend aggregate (longevity ranking, per-advertiser variation/cadence, new/disappeared) and, when provided, the ad-pattern copy appeals. Narrates which ads are validated by longevity, who pumps out variations, what is tested vs
competitor-curatorPer-persona precision selection. Takes the discovery-scout candidate pool + user seeds + the persona definition, scores persona/JTBD fit, ranks, dedupes, and proposes a per-persona shortlist. Integrates user confirmation (HARD GATE) and persists the confirmed set to the persona's competitors node. U
copy-analystAnalyzes ad COPY from an perception artifact's text CONTENT ONLY — classifies each text into a role (headline/subcopy/cta/badge/price/review_quote/spec_label), the hook type (question/contrast/result/empathy/number), sentence patterns, and keywords (feeds the keyword model). Ignores coordinates/font
copy-layout-plannerGenerates per-candidate Korean copy (headline/subcopy/CTA) and the layout plan (positions, text density) for each selected ad format, using the projected persona, product USP, claim constraints, and copy/layout principles. Use after the creative brief, before the image-prompt adapter.
creative-brief-analystSynthesizes a creative brief for image-generation by combining the projected brand/product/persona, the creative-opportunity (the precomputed strategic position + gap/whitespace + brief_constraints), review evidence summary, abstracted competitor/category patterns, and selected global principles. Us
creative-opportunity-mapperSelects strategic creative positions for OUR next ad from the market-position matrix — the ANALYSIS→GENERATION bridge (ring 3, where our product's selling-point + persona fit enter). Consumes the benefit×funnel matrix (crowded/whitespace) and emits selected_opportunities with brief_constraints the c
critic-verifierFinal gate for image-generation candidates. Critiques overclaiming, weak evidence, duplicate candidates, brand mismatch, and missing verification fields, then confirms each candidate's verification checklist. Use after adapter outputs exist, before presenting candidates. Read-only verdict.
direction-verifierBrief-boundary direction gate for image-prompt generation. Verifies the creative brief still aims at the confirmed ad_intent + persona and builds on the selected opportunity, BEFORE copy-layout spends tokens on it. Use right after creative-brief-analyst, before copy-layout-planner. Emits a verdict w
discovery-scoutRecall-maximizing competitor/advertiser discovery for ONE (product, persona). Searches public ad-transparency libraries cheaply (search and list only — no detail expand, no reviews, no image download) and returns a broad candidate pool with coverage flags. meta_ad_library and google_ads_transparency
image-prompt-adapterConverts a provider-neutral CreativeCandidateSpec into ChatGPT and Gemini image-prompt artifacts (prompt, negative_prompt, provider_notes, input_assets, expected_output, verification_checklist, retry_instruction_template). Prompt-only — never calls a real image provider. Use after copy/layout, per c
intent-analystReads WHY an ad is built the way it is — its persuasion strategy (axis 5) and the MEANING of its copy×layout bindings (axis 6). TEXT-ONLY: derived from copy-analysis + layout-analysis + visual-analysis + bindings, never the image. Ring 2, brand-free: it classifies what mechanism the ad USES (the app
interview-controllerDrives the criteria-driven interview loop. Given the highest-priority blocker from request-evaluation, formulates ONE easy-to-answer blocker-resolution question. Not question-count based. Use whenever request-evaluation reports ready=false. Read-only; the answer is later structured by the user-answe
keyword-plannerExpands ONE (product, persona) into a broad 3-axis ad-search keyword plan (needs / use-case / adjacency) for Track-1 keyword collection. Generation only — no CDP, no collection. Emits keyword-plan.json + a per-axis announce. Use at the front of data-collection, before run-flow.mjs --from-keyword-pla
layout-analystAnalyzes ad layout from an perception artifact's geometry only (bbox, font scale, placement) — composition_type, focal point, visual hierarchy, text density, and comfort signals (crowding, whitespace, breathing room, awkward placement, balance). Does not read text meaning. Use after perception-extra
market-context-researcherBuilds a sourced external context calendar for a brand/category/date range in creative-change-analysis. Receives only market/date inputs and writes context-calendar.json; never receives creative diff/candidates and never connects events to ad changes.
orchestratorEntry point / orchestrator loop for the marketing-img system. Use when the user asks to create ad image prompt candidates, set up a brand/product/persona, collect source data, or run any marketing-img mode. Drives request-eval → interview → mode dispatch with role-scoped subagents.
pattern-synthesizerWrites the interpretive narrative of a per-persona ad-pattern ON TOP OF the deterministic aggregates (composition top-k, text-role distribution, comfort averages). Produces a concise synthesis string describing the persona's dominant layout and text-placement patterns and its comfort level. Use last
perception-extractorThe ONE vision pass of the analysis pipeline. LITERAL observation of ONE ad image into a standardized dataset — geometry+text (axes 1-2) AND scene+look (axes 3-4): text content + bbox + font scale + color + shadow; graphic placement + kind; canvas; medium; what-is-depicted (subjects/scene/framing) +
request-evaluatorClassifies a user request into a mode, detects required slots and missing/insufficient/hard-blocker slots, and decides whether the mode is ready to execute. Use at the start of every request and after each interview answer, before any mode runs. Read-only; never executes a mode.
strategy-projectorProjects a COMPLETED per-ad analysis into practical marketing-strategy dimensions — benefit_vector (purchase reason), funnel_intent (buyer stage), first_cognition (does it communicate), customer_language, audience_read, generation_reusability. TEXT-ONLY (reads analysis artifacts + the ad's own adver
temporal-change-analystInterprets deterministic creative-change candidates for one persona snapshot edge. Writes interpreted-change-events.json and creative-change-report.json from creative-diff/change-candidates plus optional context, without recomputing numbers, reopening images, claiming causality, or claiming performa
visual-analystClassifies an ad's VISUAL semantics + NAMES its register/mood from a perception artifact — TEXT-ONLY (reads perception.json, NEVER the image). Buckets perception's literal scene/look into controlled vocab (setting, product_state, prop_density, palette) and names the register (clean_minimal/premium_r
mruns: node ${CLAUDE_PLUGIN_ROOT}/shared/tools/mcp-bootstrap.mjs
node
node
Build, preview, and test UI components with Storybook.
7 skills
Statically analyze a local JS/Electron app you're authorized to study, with file+line evidence, and build an original reimplementation of what you find. Experimental. Phase 1: JS/Electron static analysis. Phase 2: native ELF/PE/Mach-O container inspection. Phase 3: .NET assembly metadata. Phase 4: EVM bytecode inventory. Phase 5: CUDA fatbin, cubin and PTX inventory. Parses, never executes. Dynamic analysis is a future phase, not yet implemented.
1 skill · 1 command · 1 agent
Agent skills for Azure SQL Developer, the Azure SQL Database engine running locally in a container. Teaches your agent to run the engine, connect, migrate, scaffold, build RAG, wire CI, and go local-to-cloud, using the real Private Preview image instead of the SQL Server image.
17 skills
Static lint for Claude Code plugins and skills, with experimental LLM scoring for skills
1 skill · 3 commands · 2 agents
S&P Global - Financial data and analytics skills including company tearsheets, earnings previews, and transaction summaries
3 skills · 1 MCP server
Ask Claude about the contractors you pay through Wingspan, what you owe them, and what is holding up a payment, then onboard contractors and log payments as drafts, with a preview before anything is written. Requires a Wingspan account; bundles the Wingspan MCP server (OAuth on first use).
8 skills · 1 MCP server
This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.