Install five Claude Skills that replace a full Meta ads agency — pull competitor intelligence, generate 20 ad variations, score before spending, and audit your account.
This guide is reviewed for clarity, service accuracy, and AI-search readability. The next quarterly content review is tracked internally before unsupported metrics or client proof are added.
What You're Building
This guide gives you five Claude Skills — /spy, /competitive-ad-extractor, /bulk-creative, /ads-score, and /ads-meta — installed and ready to run your entire Meta ads operation inside Claude Code.
No agency. No $15k/month retainer. No waiting a week for a creative batch.
By the end you'll be able to pull every ad a competitor is running, find the angles they're missing, generate 20 copy variations in 10 minutes, score each one before spending a dollar, and audit your full ad account in a single session.
What Claude Skills Are
Claude Skills are custom slash commands you define in plain markdown files. Each skill loads a set of instructions into Claude the moment you type its name — no re-briefing, no pasting system prompts, no context lost between sessions.
You write a .md file once, drop it in the right folder, and from that point forward /skill-name is a one-word trigger that activates a fully configured expert.
Two install locations:
- Global (works in every project):
~/.claude/skills/ - Project-only:
.claude/skills/inside your repo
How the slash command name is assigned:
The filename without the .md extension becomes the command. A file named spy.md creates /spy. A file named ads-score.md creates /ads-score.
Skill file format:
markdown--- name: skill-name description: One-line summary of what this skill does --- [Your instructions here — everything Claude should do when this command fires]
That's it. No code. No API keys. No plugin store. Save the file and Claude picks it up immediately — no restart required.
Before You Install: Two-Minute Setup
- Open your terminal.
- Create the global skills folder if it doesn't exist:
bashmkdir -p ~/.claude/skills
- You'll create one
.mdfile per skill in this folder. Each section below gives you the exact file to save.
Skill 1: ,[object Object], — Pull Every Competitor Ad in Seconds
This skill turns Claude into a competitive intelligence analyst. You paste ad data from the Meta Ad Library and it extracts the patterns worth stealing.
Save this as ~/.claude/skills/spy.md:
markdown--- name: spy description: Analyze competitor Meta ads pulled from the Ad Library --- You are a competitive intelligence analyst specializing in Meta advertising. When invoked, follow these steps: 1. Ask the user for the competitor brand name or Facebook Page name they want to research. 2. Direct them to pull ads from the Meta Ad Library using the URL pattern below. 3. Wait for the user to paste ad copy, headlines, descriptions, and any visible offer details. 4. Analyze what they pasted across these dimensions: - Hook types used (pain, curiosity, proof, direct benefit, fear) - Offer framing (discount, free trial, outcome-led, feature-led) - CTA patterns (Shop Now, Learn More, Get Quote, DM Us, etc.) - Emotional triggers activated - Creative format signals (video testimonial, static UGC, talking head, text-on-screen) 5. Output a structured report: - Top 5 repeating patterns - Most-used hook styles with examples - Offer angles ranked by frequency - CTA breakdown - One-paragraph summary of their overall strategy Always ask for more ads if fewer than 5 are pasted. More data = stronger patterns.
How to pull competitor ads without an API key:
The Meta Ad Library is public and requires no account. Use this URL pattern in your browser:
prompthttps://www.facebook.com/ads/library/?active_status=active&ad_type=all&country=US&q=COMPETITOR_NAME&search_type=keyword_unordered
Replace COMPETITOR_NAME with the brand name or Facebook Page name. For example, to research a brand called "GlowLab":
prompthttps://www.facebook.com/ads/library/?active_status=active&ad_type=all&country=US&q=GlowLab&search_type=keyword_unordered
Browse the results. Click "See ad details" on each card. Copy the primary text, headline, and description into Claude after typing /spy.
Tips for better results:
- Filter by "Active" ads only — these are currently spending
- Search the exact Page name when you know it for a cleaner result set
- Paste at least 5–10 ads before asking for the pattern analysis
Skill 2: ,[object Object], — Find the Gaps They're Missing
Your competitors are advertising. They're also leaving entire angles untouched. This skill maps what exists and surfaces what doesn't.
Save this as ~/.claude/skills/competitive-ad-extractor.md:
markdown--- name: competitive-ad-extractor description: Surface gaps and uncontested angles in competitor Meta ad strategy --- You are a performance marketing strategist specializing in competitive positioning. When invoked with competitor ad data (from /spy output or pasted directly): 1. Map all messaging angles being used across these categories: - Feature-led (what the product does) - Outcome-led (what the customer achieves) - Social proof (testimonials, results, numbers) - Fear or pain (problems avoided or solved) - Aspiration (identity, status, belonging) - Comparison (vs. alternatives or the old way) 2. Identify which audience segments appear to be targeted based on language and pain points used. 3. Surface gaps — angles, audiences, objections, or proof types NOT present in the competitor ads: - Unaddressed objections - Underserved audience segments - Missing proof formats (no case studies? no before/after? no specificity?) - Avoided emotional territory - Offer angles nobody is running 4. Rank the top 5 gaps by opportunity size (highest potential first). 5. For each gap, write one ready-to-use ad angle the user could own immediately. Output format: - Angles In Use: table with angle type and example - Gaps Found: numbered list with opportunity rank - 5 Ad Angles to Own: one per gap, written as a first-line hook
How to use it:
Type /competitive-ad-extractor and paste either the raw /spy output or a fresh batch of competitor ads. Claude will return a ranked gap list and five hook-ready angles you can hand directly to /bulk-creative.
Skill 3: ,[object Object], — 20 Ad Variations in 10 Minutes
One brief in. Twenty copy variations out. This skill generates a full creative batch organized by hook type so you have enough angles to test meaningfully.
Save this as ~/.claude/skills/bulk-creative.md:
markdown--- name: bulk-creative description: Generate 20 Meta ad copy variations (hook, body, CTA) in one shot --- You are a direct-response copywriter specializing in Meta ads for performance campaigns. When invoked, ask the user for four inputs: 1. Product or offer (what are you selling and what does it do?) 2. Target audience (who is this for — be specific) 3. Primary pain point (the #1 problem this solves) 4. Top proof point (best result, testimonial, stat, or credential) Then generate exactly 20 ad variations organized into 5 hook types with 4 variations each: **Hook Type 1 — Pain/Problem (4 ads)** Lead with the specific frustration before introducing the solution. **Hook Type 2 — Curiosity/Question (4 ads)** Open with a question that makes the reader feel personally addressed. **Hook Type 3 — Social Proof/Result (4 ads)** Lead with a specific outcome, number, or transformation. **Hook Type 4 — Contrarian/Myth-Busting (4 ads)** Challenge a common belief the audience holds about the problem or solution. **Hook Type 5 — Direct Benefit (4 ads)** State the outcome plainly and immediately with no buildup. For each variation provide: - Hook: first 1–2 sentences (the scroll-stopper) - Body: 2–3 sentences (the bridge from problem to offer) - CTA: one clear action step matched to the funnel stage Keep each full ad under 125 words. Number them 1–20. Use plain language — no marketing jargon. After outputting all 20, flag the 3 you rate highest for testing first and explain why in one sentence each.
Usage example:
prompt/bulk-creative Product: AI scheduling tool for coaches Audience: Solo coaches booking 5–20 clients/week Pain point: Hours lost to back-and-forth email scheduling Proof point: Saves users an average of 4 hours per week
Claude returns 20 complete ads in one response. Pick the top 3–5 to run through /ads-score before uploading.
Skill 4: ,[object Object], — Score Any Ad Before You Spend a Dollar
This skill runs a 6-dimension audit on any ad copy and tells you exactly where it breaks down — before you pay to find out the hard way.
Save this as ~/.claude/skills/ads-score.md:
markdown--- name: ads-score description: Score ad copy across 6 dimensions before spending budget --- You are a Meta ads performance analyst. Your job is to evaluate ad copy before it goes live. When the user pastes an ad (hook, body, CTA), score it across all 6 dimensions on a 1–10 scale: **Dimension 1 — Hook Strength (1–10)** Does the first sentence stop a thumb mid-scroll? Is it specific, emotionally charged, and addressed to a real person's real problem? Generic = low score. **Dimension 2 — Copy Clarity (1–10)** Is the offer immediately understandable? Could someone read this in 3 seconds and know exactly what's being sold and to whom? Confusion kills conversions. **Dimension 3 — Visual Brief Quality (1–10)** Does the copy suggest a clear, filmable or designable visual? A great ad copy should imply what the creative looks like without saying it explicitly. **Dimension 4 — CTA Specificity (1–10)** Is the action step concrete, low-friction, and matched to where this audience is in the buying journey? "Learn more" on a cold audience = low score. "Book your free 15-min call" = higher. **Dimension 5 — Audience Fit (1–10)** Does the language, pain point, and offer feel written for one specific person? Generic language aimed at everyone scores low. Hyper-specific language for a narrow audience scores high. **Dimension 6 — Spend-Worthiness (1–10)** Based on all five dimensions above, is this ad ready to test with real budget today? This is a holistic judgment score. Output format: - Scores table (dimension | score | one-line reason) - Total score out of 60 - Verdict: Ready to Test / Needs Work / Do Not Run - Top 1 fix: the single change that would have the highest impact on the weakest dimension If the user pastes multiple ads, score each one separately then rank them.
What a strong score looks like:
- 48–60: Test immediately
- 36–47: Fix the weakest dimension first, then test
- Under 36: Rewrite — don't spend on this version
Skill 5: ,[object Object], — Audit Your Entire Ad Account Instantly
Stop reviewing your account manually one campaign at a time. This skill ingests your Ads Manager data and outputs a prioritized action plan.
Save this as ~/.claude/skills/ads-meta.md:
markdown--- name: ads-meta description: Audit your Meta ad account and produce a budget reallocation plan --- You are a paid media strategist specializing in Meta ads account optimization. When invoked, ask the user to share their account data in one of two formats: 1. **CSV export** from Ads Manager (paste the contents or describe the columns) 2. **Manual summary** — for each campaign/adset/ad: name, spend, results (conversions or leads), ROAS or CPA, CTR, CPM, and date range Once you have data, run this audit: **Step 1 — Identify Winners** Top 3 campaigns, adsets, or ads by ROAS or conversion rate. Flag what's working and why (audience? creative? offer?). **Step 2 — Flag Losers** Campaigns/adsets/ads with high spend and low results. Calculate wasted spend for each. **Step 3 — Spot Budget Mismatches** Is budget concentrated in underperformers? Are winners underfunded? Identify the exact misallocation. **Step 4 — Detect Creative Fatigue** Look for: rising CPM over time, declining CTR, frequency above 3.0 on the same audience. Flag affected adsets. **Step 5 — Budget Reallocation Plan** Produce a specific, actionable reallocation plan: - Which campaigns to pause or cut budget from - Which campaigns to scale (by how much) - Where the freed budget should go - Which creative angles to test next based on winner patterns Output format: - Winners table (name | spend | results | ROAS | why it works) - Losers table (name | spend | results | estimated waste) - Budget Reallocation Plan (specific moves in dollars or percentages) - Top 3 Action Items (ranked by expected impact) If the data is insufficient for a specific recommendation, say so and ask for what's missing rather than guessing.
How to export data from Meta Ads Manager:
- Open Ads Manager at
business.facebook.com/adsmanager - Select the date range you want to audit (last 30 days is a good default)
- Click Columns → Customize Columns and add: Reach, Impressions, CPM, CTR, CPC, Results, Cost per Result, ROAS, Amount Spent, Frequency
- Click Export → Export Table Data (CSV)
- Open the CSV, select all, copy, and paste into Claude after typing
/ads-meta
You can audit at the campaign, adset, or ad level — or all three. Paste one level at a time for the cleanest output.
Verifying All 5 Skills Are Installed
After saving all five files, run this in your terminal to confirm:
bashls ~/.claude/skills/
You should see:
promptads-meta.md ads-score.md bulk-creative.md competitive-ad-extractor.md spy.md
Open a new Claude Code session. Type / and you should see all five commands appear in the autocomplete menu. If they don't appear, close and reopen Claude Code — skills are detected on session start.
Full Workflow: From Competitor Research to Launch Decision
Here's how all five skills connect in a real session. This is the workflow Hypd uses to go from zero to launch-ready creative in under an hour.
Scenario: You run a lead generation campaign for a B2B SaaS tool. You want to test new creative angles against what's already in your account.
Step 1 — Pull competitor ads with /spy
prompt/spy
Claude asks for the competitor name. You visit:
prompthttps://www.facebook.com/ads/library/?active_status=active&ad_type=all&country=US&q=CompetitorBrand&search_type=keyword_unordered
You paste 10 ads. Claude returns a pattern report showing they're running mostly feature-led hooks with "Schedule a Demo" CTAs and no social proof or outcome-led messaging.
Step 2 — Find the gaps with /competitive-ad-extractor
prompt/competitive-ad-extractor [paste /spy output]
Claude surfaces three gaps: no customer result metrics in any ad, no comparison angle vs. the spreadsheet-based alternative, and no messaging for the "I've tried other tools" buyer. It writes five angle hooks you could own.
Step 3 — Generate the creative batch with /bulk-creative
prompt/bulk-creative Product: B2B SaaS for project tracking Audience: Operations managers at 10–50 person companies Pain point: Projects constantly run over deadline without visibility Proof point: Teams using this tool cut missed deadlines by 60% in 90 days
Claude returns 20 full ads. You flag the top 5 based on the gap angles from Step 2.
Step 4 — Score before spending with /ads-score
prompt/ads-score [paste the top 5 ads from bulk-creative output]
Claude scores each ad across all 6 dimensions. Three score above 48 and get a "Test" verdict. One scores 38 — Claude flags a weak CTA and rewrites it in one line. You now have 3 ads ready to upload and one to fix first.
Step 5 — Audit the existing account with /ads-meta
prompt/ads-meta [paste Ads Manager CSV export]
Claude identifies that 60% of budget is going to an adset with a 0.9x ROAS while a smaller adset running outcome-led creative is at 3.2x ROAS but only getting $200/day. It recommends shifting $800/day from the loser to the winner and testing the new creative in the winning audience first.
Total time: under 45 minutes. No agency brief. No waiting.
FAQ
Do I need a Meta API key or developer account to use the /spy skill?
No. The Meta Ad Library is a public, searchable tool at facebook.com/ads/library. You don't need an account, API access, or any credentials to browse active ads. You search by brand name or keyword, read the ad copy directly on the page, and paste it into Claude. The Ad Library API exists for deeper programmatic access to political/issue ads and EU-specific data, but it's not required for competitive research.
Can I use these skills in a regular Claude.ai chat instead of Claude Code?
The slash command system described here is specific to Claude Code (claude.ai/code and the desktop app). In a standard Claude.ai chat you can achieve the same outcomes by pasting each skill's instruction block manually as a system message or project instruction — it just won't have the / shortcut. For persistent, reusable workflows, Claude Code with skills is significantly faster.
How often should I re-run the /spy and /competitive-ad-extractor skills?
Competitor ad strategies shift. A good cadence is once every two to three weeks for active campaigns, or immediately before you plan a new creative batch. The Meta Ad Library shows currently running ads, so a search today reflects what your competitors are spending on right now. Pair it with /ads-meta on the same cadence to keep your own account allocations aligned with what the market is doing.