ai-skill-collection

Build, secure, and package Electron desktop apps — main/preload/renderer split, contextIsolation-safe typed IPC over a preload bridge, BrowserWindow + application menu + app naming, a strict CSP, and distribution (electron-builder/Forge, productName, code signing). Use when building an Electron app, wiring IPC, hardening security, fixing the app/menu name, or packaging. Do NOT use for web-only React UI (use cleanui) or Expo/React Native.

Write an engineering tech spec (design doc / RFC) that turns an approved PRD into an implementable design. Use when turning a PRD or epic into a design doc, writing an RFC, or answering 'how should we build this'. Do NOT use for product requirements (prd-development) or recording a single architectural decision (adr).

Run a right-sized threat-modeling pass (trust boundaries, STRIDE, abuse cases) on a forming design so security decisions land before Sign-Off 2. Use when a design adds attack surface — a new endpoint, data class, integration, or authz change. Do NOT use for code-diff security review (/security-review), pentests, or compliance audits.

Build and maintain a durable automated E2E suite — Playwright for web, Maestro for Expo/React Native: stable selectors, owned test data, CI gating, and a flake policy with teeth. Use when automating the E2E journeys a test plan defines or fixing a slow/flaky suite. Do NOT use for deciding what to test (test-strategy) or unit tests (tdd).

Run a go/no-go release review producing a scored readiness report — rollout plan, observability, rollback, comms verified before shipping. Use when preparing a release/launch, a go-live checklist, a rollout plan, or 'are we ready to ship'. Do NOT use for store-submission mechanics (expo-deployment) or writing the test plan itself (test-strategy).

Define production health before launch: user-centric SLIs, SLO targets with error budgets, burn-rate alerts that page on user pain (not CPU), and a runbook per pageable alert. Use when designing SLOs, alerts, error budgets, or runbooks. Do NOT use for the go/no-go review (release-readiness) or post-incident analysis (incident-postmortem).

Turn acceptance criteria into a right-sized test plan: unit vs integration vs E2E split, test-double policy, and the merge/release done-bar. Use when planning tests for a feature or epic, deciding the test pyramid, or making coverage decisions ('how should we test this'). Do NOT use for writing test cases in code or reviewing an existing diff.

Run a blameless postmortem: reconstruct the timeline, find contributing factors (never one root cause, never a person) and owned, dated action items. Use after an incident/outage/sev for a postmortem, incident review, 'what went wrong' or RCA. Do NOT use during an active incident (stabilize first; sentry-cli + runbooks) or for retro of normal work.

Scaffold or retrofit a project with docs-driven agent conventions: a root AGENTS.md as the single source of agent guidance, a thin CLAUDE.md importing it, an append-only LEARNINGS.md, and a docs/ tree with routing index and self-evaluating epics. Use when asked to set up AGENTS.md, agent/contributor docs, or make a repo agent-ready.

Build on-device AI in React Native and Expo apps with React Native ExecuTorch — LLM chat, vision-language, image classification, object detection, OCR, segmentation, image generation, speech-to-text, text-to-speech, embeddings. Use when the user wants offline AI, on-device ML, privacy-preserving or no-cloud inference, or local models on mobile.

Amazon Ads deep analysis: Sponsored Products, Sponsored Brands, Sponsored Display, and Amazon DSP — campaign structure, ACOS/TACOS, search-term harvesting, negatives, bids, ASIN targeting. Use when user says Amazon Ads, Amazon PPC, Sponsored Products, DSP, ACOS, TACOS, AMS, or retail media.

Apple Ads (Apple Search Ads) analysis for mobile app advertisers: campaign structure, bid health, Custom Product Pages, AdAttributionKit, MMP attribution, TAP coverage, CPA benchmarks. Use when user says Apple Ads, Apple Search Ads, ASA, App Store ads, AdAttributionKit, or an iOS app advertiser.

Cross-platform attribution health audit: AdAttributionKit, GA4 attribution models, Consent Mode V2, server-side stitching, MMP health, cross-device. Use when user says attribution audit, AdAttributionKit, AAK, view-through, conversion window, MMP audit, AppsFlyer, Adjust, Branch, or Singular.

Full multi-platform paid advertising audit via parallel subagents across Google, Meta, LinkedIn, TikTok, Microsoft, and Apple Ads, with per-platform health scores and an aggregate 0-100 score. Use when user says audit my ads, account health check, paid media audit, ad spend audit, or PPC audit.

Budget allocation and bidding strategy review across ad platforms: spend distribution, bidding fit, scaling readiness, kill/scale calls via the 70/20/10 rule, 3x Kill Rule, and 20% scaling rule. Use when user says budget allocation, bidding strategy, ad spend, ROAS target, or scaling.

Competitor ad intelligence across Google, Meta, LinkedIn, TikTok, Microsoft, and Apple Ads: copy, creative strategy, keywords, estimated spend via Meta Ad Library, Google Ads Transparency Center, TikTok Top Ads. Use when user says competitor ads, ad spy, competitor PPC, or ad intelligence.

Campaign concept and copy brief generator for paid ads: reads brand-profile.json and optional audit results, outputs campaign-brief.md. Run after /ads dna, before /ads generate. Use when user says create campaign, campaign brief, ad concepts, write ad copy, or creative brief.

Cross-platform creative quality audit: ad copy, video, image, format diversity; detects creative fatigue, scores diversity for Andromeda Entity-ID retrieval, sets production priorities. Use when user says creative audit, ad creative, creative fatigue, creative diversity score, or creative review.

Brand DNA extractor for paid advertising: scans a website URL for visual identity, tone of voice, colors, typography, imagery; outputs brand-profile.json. Run before /ads create or /ads generate. Use when user says brand DNA, brand profile, extract brand, brand identity, or brand style guide.

AI image generation for paid ad creatives: reads campaign-brief.md and brand-profile.json to produce platform-sized ad images via banana-claude (v1.4.1+ with nanobanana-mcp). Use when user says generate ads, create ad images, make ad creatives, generate visuals, or generate from brief.

Google Ads deep analysis across Search, Performance Max, AI Max, Display, YouTube, and Demand Gen: conversion tracking, wasted spend, keywords, Smart Bidding. Use when user says Google Ads, PMax, Performance Max, AI Max, AI Brief, broad match, Quality Score, search terms audit, or Smart Bidding.

Landing page quality assessment for paid campaigns: message match, page speed, mobile experience, trust signals, form optimization, conversion potential. Use when user says landing page audit, LP audit, post-click experience, landing page CRO, or conversion rate.

LinkedIn Ads deep analysis for B2B: technical setup, audience targeting, creative quality, lead gen forms, bidding, Thought Leader Ads, ABM, predictive audiences. Use when user says LinkedIn Ads, B2B ads, sponsored content, lead gen forms, InMail, ABM, Thought Leader Ads, or LinkedIn campaign.

PPC financial calculator: CPA, ROAS, CPL, break-even, impression share opportunity sizing, budget forecasting, LTV:CAC, MER. No API access needed — works with pasted export data. Use when user says PPC math, break-even, ROAS calculator, budget forecast, impression share, LTV CAC, or MER.

Meta Ads deep analysis (Facebook, Instagram, Threads): Pixel/CAPI health, creative diversity and Entity-ID clustering, ASC/AAC defaults, Advantage+ and creative-as-targeting scoring. Use when user says Meta Ads, Facebook Ads, Instagram Ads, Advantage+, ASC, AAC, Andromeda, GEM, or Lattice.

Microsoft/Bing Ads deep analysis: search, Performance Max, Audience Network, Copilot integration, Google Ads import validation, LinkedIn audience targeting, cost advantage. Use when user says Microsoft Ads, Bing Ads, Bing PPC, Copilot ads, or Google import audit.

Product photography enhancement for ad creatives via banana-claude (v1.4.1+ with nanobanana-mcp): generates 5 styles from one product image — Studio, Floating, Ingredient, In Use, Lifestyle. Use when user says product photo, photoshoot, studio shot, lifestyle photo, or enhance product image.

Server-side tracking pipeline audit: sGTM, Meta CAPI Gateway, Conversions API health, event_id deduplication, server-side hit ratio, pixel debugging, PII hashing. Use when user says server-side tracking, sGTM, CAPI, Conversions API, event deduplication, pixel health, or iOS 14.5 recovery.

A/B test design for paid ads: hypothesis framework, statistical significance calculator, test duration and sample size estimators, setup guides for Meta, Google, and LinkedIn experiments. Use when user says A/B test, split test, experiment design, statistical significance, or sample size.

TikTok Ads deep analysis: creative-first strategy, safe zone compliance, tracking, bidding, campaign structure, TikTok Shop, Smart+ campaigns, creative diversity for retrieval. Use when user says TikTok Ads, TikTok marketing, TikTok Shop, Spark Ads, Smart+, or USDS.

YouTube Ads analysis across skippable and non-skippable in-stream, bumper, Shorts, Demand Gen, and Connected TV formats: creative quality, targeting, measurement, frequency capping. Use when user says YouTube Ads, video ads, pre-roll, bumper ads, Shorts ads, Demand Gen, VAC, or CTV.

Turn raw PM content into a compliant, publish-ready skill, or design a new skill through guided Q&A that shapes its type, scope, and structure. Enforces THIS repo's conventions and validation gates (generic scaffolding is skill-creator's job). Use when creating or updating a repo skill — even from just an idea or notes.

Generate or edit images with GPT Image 2 / gpt-image-2 — text-to-image, reference-image editing, inpainting, posters, typography, UI mockups, diagrams. Search the bundled Reference Gallery, then use the `gpt-image` CLI or `scripts/generate.py`; don't write new image code. Use when asked to generate, draw, render, or edit images.

Read, search, triage, draft, and send mail plus manage calendar events in Outlook / Microsoft 365. Requires the `outlook` CLI; prefer draft-* over direct send. Use when asked to check inbox, summarise unread mail, find or reply to an email, view the agenda, create or update events, RSVP, or check free/busy. Does NOT cover Gmail or IMAP.

Diagnose hard bugs and performance regressions: build a red-capable feedback loop, reproduce, hypothesise, instrument, and fix at the root cause with a regression test. Use when debugging, hunting a root cause, asking 'why is this failing', or chasing flaky behavior. Do NOT use for post-incident reviews (incident-postmortem).

Build a high-quality MCP (Model Context Protocol) server exposing an external API or service to LLMs — TypeScript (MCP SDK) or Python (FastMCP): tool design, schemas, error handling, transport, testing, evaluations. Use when creating or extending an MCP server. Do NOT use for configuring or consuming existing MCP servers.

Priority-ranked Postgres best practices: query performance and indexing, connection management and pooling, schema design, Row-Level Security, locking and concurrency. Use when writing or reviewing SQL or Postgres schema, fixing slow queries, or planning migrations. Do NOT use for API contract design (`api-design`) or managed-DB provisioning.

Impact-ranked React/Next.js performance rules from Vercel — eliminating waterfalls, bundle size, server/client data fetching, re-renders. Use when writing or reviewing React/Next.js code for performance, re-renders, waterfalls, bundle size. Do NOT use for React Native (`react-native-skills`) or visual design (`cleanui`).

Impact-ranked React Native/Expo performance rules from Vercel — list/scroll performance, Reanimated animations, images, native modules, navigation. Use when building or optimizing React Native/Expo app code. Do NOT use for Expo APIs, builds, or deployment (the `expo-*` skills) or React web (`react-best-practices`).

Resolve an in-progress git merge or rebase conflict hunk by hunk: trace each side's original intent, preserve both where possible, verify with the project's checks, and finish the merge. Use when a merge/rebase stops on conflicts or files contain conflict markers. Do NOT use for branch strategy or general git history rewriting.