Your competitors already paid to find out what converts. Point your AI agent at the public ad libraries and turn their proof into your concepts:

  • Automate collection: guided sweeps of the Meta, TikTok, Google & LinkedIn ad libraries, triaged by what's proven by run-time.
  • Automate teardowns: scored, evidence-anchored analysis of every ad: scene by scene, with the call on what actually carries it.
  • Automate synthesis: patterns weighted by what ran longest, turned into an angle bank and production-ready concept briefs for your brand.
  • Stay clean by design: structure transfers, content never; every claim must be true for your product.

One download with your free customer account, free for commercial use. Open it with your agent and run your first collection session today.

How to set it up

  1. Download and unzip. You get a folder called ad-spy-starter: 18 plain-text files, no code, no installs, no paid spy tools.
  2. Open it with your AI agent. The workspace is agent-agnostic: Claude Code, Codex, Cursor, or whatever agentic system you already use; any agent that can read files can run it (image-capable agents tear down creatives directly). In Claude Code the included CLAUDE.md orients the agent automatically; elsewhere, say "read CLAUDE.md first".
  3. Fill in your spy profile. Open context/spy-profile.md: your brand, your differentiator, what you can truthfully claim, and the advertiser watchlist (3–5 direct competitors + 2–3 adjacent advertisers, often the richer source). Or tell the agent: "Interview me against the spy profile template."
  4. Run your first collection session. Workflow 01 walks you through the free ad libraries (Meta, TikTok Creative Center, Google Ads Transparency, LinkedIn) with the triage that matters: an ad running 6+ weeks is proven-by-spend. The agent tells you exactly what to capture per ad (the evidence kit).
  5. Let the agent tear them down. Workflow 02: the systematic per-ad analysis. Scored dimensions, scene-by-scene for video, the landing-page check, and the committed call on which element actually carries the ad.
  6. Synthesize and adapt. After 10–15 teardowns, workflow 03 turns the swipe file into your market's proven patterns, then into your angle bank: ad concepts for your brand that borrow the winning mechanics with your substance, your claims, your voice. Test, log the results, and the next round learns from them.

What this is

Your competitors' ads are public. Every major platform runs a free, open ad library, and an ad that has been running for months survived because it converts. This resource hands you a working agent workspace that turns that public evidence into an operation:

  • The collection protocol: where to look across the four libraries, the longevity triage, the rotation read (what's scaling, what just died; free intelligence most people never read), and the evidence-kit spec that makes every ad analyzable once, completely.
  • Systematic teardowns: evidence-anchored analysis the way an analyst works, not a scroller. Scored dimensions, scene-by-scene walks for video, the visual-hierarchy read for statics, the landing-page check, and a committed load-bearing call per ad.
  • Pattern synthesis: every teardown ends in a uniform synthesis block, so twenty teardowns aggregate mechanically into a longevity-weighted pattern report with an honest confidence ceiling.
  • The concept adapter: proven mechanics rebuilt for your brand through hard gates. Structure transfers, content never; every claim true for your product; every concept leaning on your differentiator. Output: production-ready concept briefs with test notes.

Why agent workflows, not a swipe folder

Because a folder of screenshots is not intelligence. These agent workflows give your agent a memory (your spy profile, your swipe log with the rotation record, your angle bank with real test results) and a contract: every analytical claim anchored to evidence, patterns weighted by what actually ran longest, and your own results outranking market proxies in the next round. That is the same architecture always-on creative-intelligence systems are built on: complete, yours, and ready to run this week.

What makes it different

  • The agent orients before acting. Every session starts with a workspace status: watchlist coverage, teardown backlog, pattern freshness, concepts in test.
  • Longevity over opinion. The free libraries date every ad. The workspace reads run-time and variant families as the performance proxy they are, and never invents metrics it cannot see.
  • The legal and ethical line is built in. Structure transfers, content never: no copied lines, no competitor assets, no claims that aren't true for your product. The agent enforces this even when it would be convenient not to.
  • It ends production-ready. Concept briefs ship with hook, beats, CTA, visual direction, and a fair-test note. If you also use our free AI Social Media workspace, its production planning takes a concept brief directly.

It's free

Free under CC-BY-4.0: use it, adapt it, and ship it commercially. Attribution appreciated.