功能描述

Competitive ad intelligence skill for fetching, analyzing, and reporting on competitor ads across Meta (Facebook/Instagram), Google Ads Transparency Center,...

使用说明 (SKILL.md)

Ad Intelligence Skill

Name: ad-intelligence-skill
Author: abhishekj9621

A two-phase competitive ad intelligence skill for marketing companies and ecommerce owners to research how competitors run ads across Meta, Google, and LinkedIn.

Quick Reference: What Data Is Available

Field	Meta (FB/IG)	Google Transparency	LinkedIn
Ad copy / headline	✅ Both phases	✅ Both phases	✅ Both phases
Creative (image/video URL)	✅ Both phases	✅ Both phases	✅ Both phases
Ad format type	✅ Both phases	✅ Both phases	✅ Both phases
Date first/last shown	✅ Both phases	✅ Both phases	✅ Both phases
Platform / placements	✅ Both phases	✅ (Google Search, YT, Display, etc.)	✅ Both phases
Active / inactive status	✅ Both phases	✅ Both phases	✅ Both phases
CTA button	✅ Phase 1 scrape	❌	✅ Both phases
Destination / landing URL	✅ Both phases	✅ Both phases	✅ Both phases
Total ads running	✅ Phase 2	✅ Phase 2	✅ Phase 2
Impression range	✅ Phase 2 (EU/political only)	❌	✅ Phase 2 (some scrapers)
Spend range	✅ Phase 2 (EU/political only)	❌	❌
Demographic breakdown	✅ Phase 2 (EU only)	❌	❌
Targeting info	❌	❌	✅ Phase 2 (partial — language, location)
Advertiser ID	✅ Phase 2	✅ Phase 2	✅ Phase 2

⚠️ No platform exposes: exact spend, CTR, conversion rates, ROAS, or detailed audience targeting.

Workflow

Step 1: Understand the Request

Gather from the user (ask if not provided):

Competitor name or domain (e.g., "Nike" or "nike.com")
Target platform(s): Meta / Google / LinkedIn (default: all three)
Phase: Phase 1 (quick, no API keys) or Phase 2 (deeper, requires API keys)
Country/region filter (optional, e.g., "US", "IN", "EU")
Date range (optional, e.g., "last 30 days")
Ad format filter (optional: image / video / carousel / text)

If the user hasn't said which phase they want, ask. If they're just exploring, start with Phase 1.

Step 2: Execute the Right Phase

Read the platform reference files for code, endpoints, and examples:

Meta: references/meta.md
Google: references/google.md
LinkedIn: references/linkedin.md

Each reference contains:

Phase 1: Python scraping code (no API key needed)
Phase 2a: Official/free API instructions
Phase 2b: Third-party paid API options (SerpAPI, SearchAPI, Adyntel, Apify)
Sample output JSON
Known limitations

Step 3: Format the Output

Always deliver both:

A. Human-Readable Summary Report

## Ad Intelligence Report: [Company Name]
**Platforms Searched:** Meta, Google, LinkedIn
**Date:** [today]
**Phase:** 1 (Scrape) / 2 (API)
**Country:** [region]

### 📊 Overview
- Total ads found: X
- Active ads: X | Inactive: X
- Formats: X% image, X% video, X% carousel

### 🎯 Platform Breakdown
[Per platform: count, date range, notable trends]

### ✍️ Creative Themes & Messaging Patterns
[Summarize recurring hooks, CTAs, offers, tone]

### 📅 Recency & Cadence
[How often they post new ads, seasonal patterns]

### ⚠️ Data Limitations
[Note what wasn't available and why]

B. Raw Structured Data (JSON)

Return normalized JSON with this schema for each ad:

{
  "platform": "meta|google|linkedin",
  "ad_id": "string",
  "advertiser_name": "string",
  "advertiser_page_url": "string",
  "status": "active|inactive",
  "format": "image|video|carousel|text|document",
  "headline": "string|null",
  "body_text": "string|null",
  "cta_text": "string|null",
  "destination_url": "string|null",
  "creative_url": "string|null",
  "platforms_served": ["facebook", "instagram", "messenger"],
  "date_first_shown": "YYYY-MM-DD|null",
  "date_last_shown": "YYYY-MM-DD|null",
  "country": "string|null",
  "spend_range": "string|null",
  "impression_range": "string|null",
  "targeting_summary": "string|null",
  "source_phase": 1
}

Step 4: Provide Next Steps

After presenting results, always suggest:

Which Phase 2 option would give more depth for this use case
What API keys/accounts are needed to upgrade
Whether the data is sufficient or the user should broaden/narrow their search

Important Limitations to Always Communicate

No private performance data: CTR, conversions, ROAS, exact spend — these are never public.
Meta API geographic restriction: The official Meta Ad Library API only returns impression/spend data for EU-delivered ads and political/social cause ads globally. For general ecommerce ads in non-EU regions, Phase 1 scraping often gives broader coverage.
Google has no official public API: All Google Transparency Center data is accessed via scraping (Phase 1) or third-party wrappers like SerpAPI/SearchAPI (Phase 2).
LinkedIn targeting data is partial: Only language and location hints are sometimes visible — not job title, seniority, or company size targeting.
Rate limits apply: All platforms rate-limit requests. Use pagination and add delays (1–2s between calls).

Phase Decision Guide

Situation	Recommended Phase
Quick competitive overview	Phase 1
No API keys available	Phase 1
Need date-range filtering	Phase 2
EU market + want spend/impressions	Phase 2 (official Meta API)
Need 500+ ads at scale	Phase 2 (paid third-party API)
B2B competitor research on LinkedIn	Phase 2
Need creative image/video downloads	Phase 2

Reference Files

references/meta.md — Meta (Facebook/Instagram) scraping + API code
references/google.md — Google Ads Transparency scraping + SerpAPI/SearchAPI
references/linkedin.md — LinkedIn Ad Library scraping + third-party APIs

安全使用建议

This skill appears to do what it says (collect ad creatives via scraping and paid APIs), but there are several red flags to resolve before installing or giving it secrets: 1) Metadata omits required tools and secrets — Phase 1/2 code needs Python packages and Playwright/browser binaries, and Phase 2 needs multiple API keys (SERPAPI, Apify, Adyntel, ScrapeCreators, Meta Graph access token). Ask the skill author to declare required env vars and an install spec. 2) Do not paste high-privilege tokens (long-lived Facebook access tokens, service API keys) into the agent until you confirm where they are stored and how they will be used; prefer creating scoped keys with minimal permissions. 3) Scraping reverse-engineered internal APIs (Google internal endpoints) and headless browser automation can violate platform TOS and lead to IP blocking or account actions — consider legal/terms-of-service risk. 4) Run this skill in an isolated environment (sandbox) if you need to test, and limit network/credential exposure (use limited-scope API keys, proxies, and monitoring). 5) Request from the publisher: an explicit install section listing required packages/binaries, a clear list of all environment variables the skill will request, and a data-handling statement describing whether/where keys or scraped data are transmitted or stored. Only proceed after those clarifications and after minimizing the privileges of any provided credentials.

能力评估

ℹ Purpose & Capability

The name and description (fetch/analyze competitor ads across Meta, Google, LinkedIn) align with the provided reference files and scraping/API code examples. The requested capabilities (scraping ad libraries, calling third-party ad-intel APIs) are coherent with the stated purpose.

⚠ Instruction Scope

SKILL.md and the reference files explicitly instruct the agent to run web scrapers and headless browsers (Playwright/Selenium), call internal/reverse-engineered endpoints (Google internal API), intercept network responses, and use many third-party APIs. The instructions reference local runtime actions (browser automation, request interception), and expect the agent to accept API keys from the user for Phase 2 — but the skill metadata does not declare these requirements. The scope is broad and includes activities (intercepting network responses, running headless browsers) that require specific binaries and permissions not declared.

⚠ Install Mechanism

There is no install specification, yet the references rely on Python packages (Google-Ads-Transparency-Scraper, serpapi, apify_client, requests, playwright), Playwright browser installation, and possibly system-level browsers/proxies. That mismatch (no install steps vs. explicit tooling requirements) is an operational and security concern because users/agents won't be told what will be installed or what prerequisites are required.

⚠ Credentials

The registry lists no required environment variables or primary credential, but the instructions and reference code expect many secrets for Phase 2: SERPAPI_KEY (or env fallback), Apify tokens, Adyntel API keys, ScrapeCreators keys, Meta Graph API access tokens, and possibly other third‑party credentials. These are proportionate to Phase 2 functionality but their absence from the declared metadata is a mismatch and increases risk (agents/users may be prompted to paste sensitive keys without clear justification or constraints).

✓ Persistence & Privilege

The skill is not always-enabled and does not request system-wide persistence. There is no install script in the registry that writes or modifies agent configs. Autonomous invocation is allowed (default) but not combined here with other high-risk flags.

版本历史

v1.0.0

- Initial release of the Ad Intelligence Skill, providing competitive ad research across Meta, Google, and LinkedIn. - Supports two phases: Phase 1 via web scraping (no API keys), Phase 2 using official and third-party APIs for deeper insights. - Returns both a human-readable summary report and structured JSON data for all ads found. - Guides users through required input (competitor, platform, phase, filters) and recommends optimal workflow for different use cases. - Clearly communicates platform data limitations, phase capabilities, and necessary next steps for advanced research.

元数据

Slug ad-intelligence-skill

版本 1.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题