Meta Ads logo

Meta Ads

Community
boringmarketer
meta-ads

Build and verify Meta (Facebook/Instagram) ad campaigns end to end via the Marketing API — account/app/system-user permission chain, Special Ad Category constraints, crop-safe story creative, campaign creation, and conversion tracking that is actually proven to work. Use when launching or debugging Meta campaigns, Meta pixel, or Conversions API. Encodes the exact API rejections, permission gotchas and false-negative verification traps hit in production.

Overview

Publisherboringmarketer
Repositorykimi-first
Skill namemeta-ads
Stars
51
Forks
2
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by boringmarketer on GitHub. Read the source before you install it.

Installation

Install the Meta Ads AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/boringmarketer/kimi-first/tree/main/skills/meta-ads
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/boringmarketer/kimi-first.git /tmp/kimi-first
mkdir -p .claude/skills
cp -r /tmp/kimi-first/skills/meta-ads .claude/skills/meta-ads
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Meta Ads in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Meta Ads on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Meta Ads is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Meta Ads — launch and verification

Everything here was learned by getting it wrong in production first. The API rejections are real rejections, the permission chain is the one that actually unblocks, and the verification section exists because measuring the wrong thing produced three confidently-wrong conclusions in a single session.


1. Verification comes first, because it is where the damage happens

Read this before believing any test result.

Three separate false negatives in one launch. Each looked like a real bug. Each was the instrument, not the system.

TrapWhat happenedWhat to use instead
Testing on localhostVerified the pixel and an embed on localhost. Both were blocked in production by a CSP that localhost does not have.Always verify against the deployed URL. A localhost pass proves nothing about production headers.
performance.getEntriesByType('resource')Concluded "the Lead never fires" because no beacon appeared. fbevents uses navigator.sendBeacon when an eventID is present, and sendBeacon never appears in resource timing.Meta's Test Events tool. It is Meta's own instrumentation and it ends the argument.
"Received From: Browser" in Test EventsConcluded the server-side CAPI was broken because no server rows appeared. Test Events only shows server events if the server sends a test_event_code.Have the server return its own send result (see §6). Absence of display ≠ absence of delivery.

The pattern: absence of evidence was read as evidence of absence, from an instrument that structurally could not observe the thing being measured. Before concluding something is broken, ask what the instrument can actually see.

Confounded variables. One "isolation test" removed two variables at once and the wrong one got the blame. Change one thing.


2. The permission chain (the part that actually blocks you)

To create ads via API you need four separate grants. Missing any one gives a different, unhelpful error.

  1. A Meta app — you almost certainly already have one; don't create a new one.
  2. The system user has a ROLE ON THE APP. Business Settings → Apps → app → Assign people → pick the system user → Develop app. Skipping this gives No permissions available — assign an app role to the system user at the token-generation step, with no hint about which grant.
  3. The system user has the ASSETS. System Users → user → Assign assets:
    • Ad account → Manage campaigns (not "Manage ad accounts", which is finances and permissions)
    • Page → Ads only
    • Pixel/dataset → Use events dataset
  4. Generate the token — System Users → Generate token → app → Never expiry → tick ads_management.

Two more that only surface at ad-creation time:

  • Instagram account connected to the ad account. Advantage+ placements include IG, and creation fails with "Ad account has no access to this Instagram account". Fix: Business Settings → Instagram accounts → account → Connect assets → the ad account.
  • The app must be Live, not Development. "Ads creative post was created by an app that is in development mode". Publishing needs a Privacy Policy URL and a Category, then Publish. Development mode blocks creative creation only — the rest of the API works, so this fails late.

Expiry: prefer Never for a system-user token you control. A time-boxed token lapses silently and takes the integration with it.


3. Special Ad Category (Employment, Housing, Credit)

Employment strips the targeting you would normally rely on:

  • no age or gender targeting — Meta forces 18–65, all genders
  • no detailed interest or behaviour targeting
  • minimum 15-mile radius on location targeting
  • no earnings claims anywhere — "$2k/week", "/hr", "$X per job" are rejection risk even when true

Consequence: the creative IS the targeting. With no interest targeting, the only thing making the right person recognise themselves is the ad itself. Put the audience identifier in the image as literal words — SAVANNAH DIESEL MECHANICS — not a clever hook.

Guard the money rule in code, in every place copy can be edited, and fail loudly:

js
const MONEY = /\$\s?\d|\b\d+\s?(k|dollars|usd)\b|\bper hour\b|\/hr\b/i;

Avoid "now hiring" for contractor work — it pulls W2 job seekers who click, fail qualification, and cost money on the way through.


4. Creative: the crop-safe zone

A 1080×1920 asset is cropped for feed placements. 4:5 removes ~285px from each end; 1:1 removes ~420px. Anything near the top or bottom is destroyed in exactly the placement where most impressions land.

Keep every critical element inside the 1:1 safe band: y 420 → 1500. Stack the identifier, caption and CTA as one block centred in that band, so the whole message survives any crop.

Verify by simulating the crops rather than trusting the full-size render:

js
for (const h of [1350, 1080]) {
  await sharp(file).extract({ left: 0, top: (1920 - h) / 2, width: 1080, height: h })
    .toFile(`crop_${h}.jpg`);
}

Then look at the ad preview in Ads Manager, which renders each placement for real. That is what caught the cropped identifier.

Native caption look: IG/TikTok put a rounded box behind each individual line, not one flat band — a single band reads as a slide. Heavy bold-italic in a system sans, not a brand face. No fake platform chrome (progress bars, reply boxes, swipe-up arrows) — Meta rejects creative that mimics its UI, but a plain caption band is fine.

Auto-fit rather than trusting copy length. SVG cannot size a rect to its own text, so estimate width and shrink until it fits. Italic overhangs its advance width — widen horizontal padding when italic is on.

Opt out of Advantage+ creative enhancements. They auto-crop and overlay text. The API field standard_enhancements is deprecated with a moving per-feature replacement, so set it in the UI. Check degrees_of_freedom_spec on the creative to confirm advantage_plus_creative: OPT_OUT.


5. Campaign creation — the rejections, in order

Each of these is a real 400 that stops creation. Fix them up front.

js
// CAMPAIGN
{
  objective: 'OUTCOME_LEADS',
  status: 'PAUSED',
  special_ad_categories: ['EMPLOYMENT'],
  buying_type: 'AUCTION',
  is_adset_budget_sharing_enabled: false,  // required when budget is on the ad set
}

// AD SET
{
  daily_budget: 3000,                       // CENTS
  billing_event: 'IMPRESSIONS',
  bid_strategy: 'LOWEST_COST_WITHOUT_CAP',  // required; a cap throttles a cold pixel
  optimization_goal: 'OFFSITE_CONVERSIONS',
  promoted_object: { pixel_id, custom_event_type: 'LEAD' },
  targeting: {
    geo_locations: { custom_locations: [{ latitude, longitude, radius: 25, distance_unit: 'mile' }] },
    age_min: 18, age_max: 65, genders: [1, 2],
  },
}

Errors and their causes:

ErrorCause
must specify True or False in is_adset_budget_sharing_enabledbudget on ad set, flag absent
Bid amount or bid constraints requiredno explicit bid_strategy
daily budget must be greater than $0budget arrived as null — see below
Ad account has no access to this Instagram accountIG not connected to ad account
created by an app that is in development modeapp not published
standard_enhancements has been deprecateddrop the field, set it in the UI

The null budget was a caller bug worth remembering: argv[argv.indexOf('--flag') + 1] returns argv[0] when the flag is absent, because indexOf gives -1. Number('--apply') is NaN, which JSON.stringify turns into null. Guard the flag's presence and range-check before sending.

Always create PAUSED, verify, then enable deliberately.

Clean up orphans. A failed run leaves a campaign or ad set behind; delete it before retrying or you accumulate duplicates.


6. Conversion tracking that is actually proven

Both paths, one event id. Browser pixel and server CAPI send the same event_id; Meta collapses them into one conversion. The server path survives ad blockers — roughly a third of an audience — so it is not optional.

The browser must forward what the server cannot know:

js
meta: {
  event_id: evId,
  event_source_url: window.location.href,
  fbp: cookie('_fbp'),   // first-party cookies on YOUR domain; the function
  fbc: cookie('_fbc'),   // runs elsewhere and never receives them otherwise
}

Make the server report its own outcome. A CAPI call is correctly non-fatal — losing an attribution event must never fail a form submission — which means it can rot silently forever. Return the result:

ts
return jsonResponse({ success: true, id, meta_capi: result.ok ? `sent:${n}` : `failed:${err}` })

One curl then proves the whole server path in isolation. This is the only thing that definitively settled it.

Fire events on what you mean. A Schedule fired when a calendar renders means "someone saw a calendar", not "someone booked" — optimising toward it trains Meta to find people who arrive and leave. Use the embed's own success callback (bookingSuccessful for Cal.com) and make the rendering a custom event.

Optimise for volume, measure on truth. Meta needs ~50 conversions/week per ad set to leave learning. If the true outcome is rarer than that, bid on the higher-volume upstream event and keep the real one for reporting.

Watch for automatic events. Meta's automatic event detection invents events (e.g. Subscribe) from form interactions. They can pollute optimisation — disable in pixel settings.


7. CSP will silently break all of it

If the site sends a Content-Security-Policy, these must be allowed or the pixel and any embed die with no visible error:

script-src   https://connect.facebook.net  https://app.cal.com
connect-src  https://www.facebook.com  https://connect.facebook.net
frame-src    https://www.facebook.com   (+ any embed origin)

Blocked fbevents.js leaves fbq as the inline stub: it queues events and sends nothing, forever.

The check that detects it:

js
typeof window.fbq.callMethod === 'function'   // real library loaded
window.fbq.queue.length                        // >0 and growing = stub, blocked

fbq.loaded and fbq.version are set by the inline stub and prove nothing.


Launch checklist

  • System user has app role, ad account, Page, pixel
  • App published (not Development)
  • Instagram connected to the ad account
  • Token ads_management, Never expiry
  • Copy contains no money figures — guarded in code
  • Creative content inside y 420–1500; crops simulated
  • Advantage+ creative enhancements OPT_OUT
  • Created PAUSED; ad previews checked in Ads Manager
  • fbq.callMethod is a function on the deployed site
  • Lead confirmed Processed in Test Events
  • Server CAPI confirmed sent:N via its own response
  • Test records deleted from the database afterwards
  • Token revoked/rotated if it appeared in any transcript or log

Frequently asked questions

What does the Meta Ads AI skill do?

Build and verify Meta (Facebook/Instagram) ad campaigns end to end via the Marketing API — account/app/system-user permission chain, Special Ad Category constraints, crop-safe story creative, campaign creation, and conversion tracking that is actually proven to work. Use when launching or debugging Meta campaigns, Meta pixel, or Conversions API. Encodes the exact API rejections, permission gotchas and false-negative verification traps hit in production.

Why use Meta Ads on TypingMind?

Because you install it once and use it with any model. Meta Ads is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Meta Ads in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/boringmarketer/kimi-first/tree/main/skills/meta-ads. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Meta Ads?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Meta Ads?

As many as you like. As long as a model supports skills, you can use Meta Ads with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Meta Ads AI skill free?

Yes. It is published on GitHub by boringmarketer under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇