For Agents · MCP

Your task.
The right skill.

The tunnel uses the same cards as the catalogue. Browse only as deep as needed — or load a broad bundle immediately.

No local installationNo M11 loginRead-only

Broad task

Load 5–10 and go

SEO, Sales, Agents or another broad area → one bundle call → work.

MCP endpoint: https://skills.m11.ch/mcp

Read-only access to published skills. Default 8, maximum 10 skills / 120,000 characters.

Task Packs

Marketing Foundation

Compact two-skill starter: clarify positioning and choose a lead magnet. Use Marketing Launch for the broader eight-skill go-to-market workflow.

Marketing Launch

Build an evidence-led marketing plan from ICP and competition through positioning, campaigns, growth and measurement.

Agent Audit Essentials

Diagnose architecture and context, plan agent-team responsibilities, then organize project context and session handoffs. Memory and cost-runtime reviews remain outside this pack.

Conversion and Activation Review

Review the journey from landing page and lead capture through registration, first value and transparent upgrades.

Campaign Content and Brand Review

Plan a campaign, draft its channel content and review the work against actual brand guidance.

Content and Launch Planning

Prioritize an editorial roadmap and plan how to launch and distribute it across suitable channels.

Customer, Market and Community Research

Understand customer needs, compare competitors and plan a community around real member value.

Lead Magnet to Lifecycle Nurture

Choose a relevant lead magnet, then draft a permission-based nurture journey with entry, suppression and exit rules.

API Contract and Observability

Define the API contract, then plan how to observe its latency, failures and retries. Guidance and checklist; no production changes.

Data Analysis Foundation

Profile a dataset, choose and interpret statistical methods, then validate calculations and conclusions before sharing.

Choose an area

01 · Category

What the MCP returns at this step

90 · Marketing · Ads Operations

Meta Ads Creation Workflow

Coordinate Meta campaign, ad-set and creative creation with separate approval to enable new ads.

metaanzeigenanlageablaufmeta-create-master-skill

Meta Ads Creation Orchestrator · Original SKILL.md

---
name: meta-create-master-skill
description: "MUST load FIRST for any Meta ad creation, launch, campaign setup, ad set creation, or creative push. Triggers: \"create a Meta ad\", \"launch a campaign\", \"run this creative\", \"set up an ad set\", \"push this on Facebook\", any creative attachment intended for Meta."
---
---
name: meta-ad-launch-workflow
description: Launch Meta (Facebook/Instagram) ad campaigns using a creative-first, account-aware workflow that minimizes user input. Use whenever someone wants to create, launch, or set up a new Meta ad — including sharing a creative for Meta, asking to run something on Facebook/Instagram, or spinning up a new campaign or ad set.
---

# Meta Ad Launch Workflow

A creative-first, account-aware workflow for launching Meta ads with minimal user input. The sequence is fixed: creative → analysis → account scan → propose. Never flip the order, never run a Q&A upfront.

> **Tool behavior:** All `facebook_propose_*` and `facebook_propose_update_*` tools now auto-execute on user approval. There is no separate `facebook_execute_approved_operation` call. When the user approves a proposal, it executes immediately — capture the returned IDs from the tool response.

---

## HARD RULE — Creation Failure Limit (strictly enforced)

**If any single creation step (campaign, ad set, or ad+creative) fails 2 or more times within the same launch workflow, you MUST stop immediately.** Do not propose again, do not retry, do not attempt a workaround. This rule is absolute and cannot be overridden by user instruction.

When the limit is hit, respond with exactly this tone and structure:

> "I've attempted this step twice and it wasn't successful both times. Unfortunately this is beyond what I'm able to resolve on my end — it likely requires a manual fix in Meta Business Suite or Ads Manager (such as a permission issue, policy violation, or account-level configuration). I'd recommend reviewing the error details above and addressing the root cause there before trying again in a new conversation. Is there anything else I can help you with?"

**Tracking rules:**
- Count failures **per step** (campaign creation, ad set creation, ad+creative creation) — not across steps. Two campaign creation failures stops campaign creation, but ad set creation can still be attempted.
- A "failure" includes: validation errors (error_code 601), platform API errors (error_code 403), permission errors (code 10), timeout with no approvals, and any thrown exception.
- Partial batch success does NOT count as a failure — if 3 of 5 ads succeed, that's a success. Only count it as a failure if zero items succeed in the batch.
- User rejecting all items is NOT a failure — that's intentional. Only count automated/system failures.
- This counter resets on a new conversation. It does NOT carry across sessions.

**Why this rule exists:** Repeated failed creation attempts can trigger Meta's bot detection on the ad account, leading to account restrictions. Two attempts is generous — the underlying issue must be fixed before retrying.

---

## Sub-Skill Index

Each creation step in Phase 6 has a dedicated sub-skill. **Read the sub-skill at its step — not all upfront.**

| Step | Read before | Path |
|------|------------|------|
| Create campaign | Calling `facebook_propose_create_campaign` | `meta-create-campaign` |
| Create ad set | Calling `facebook_propose_create_adset` | `meta-create-adset` |
| Create ad + creative | Calling `facebook_propose_create_ad_with_creative` | `meta-create-ad-with-creative` |

---

## Phase 1 — Get the Creative

The creative comes first. Always.

- If the user attached an image, video, or carousel — proceed to Phase 2.
- If they referenced an existing ad — pull it from the account and proceed.
- If nothing is attached — ask exactly one question: *"Share the creative, or point me to an existing ad to clone."* Nothing else. No objective, no budget, no audience questions.

---

## Phase 2 — Analyze the Creative

Extract the following before touching any account data:

- **Format** — image / video / carousel; aspect ratio
- **Hook style** — bold claim, UGC, problem-agitation, offer-led, curiosity
- **Value prop** — the core promise the viewer receives
- **CTA** — what action the creative drives (shop, sign up, learn more, get quote)
- **Audience signals** — demographic cues, lifestyle markers, occasion, problem context
- **Funnel stage** — TOF (cold awareness), MOF (consideration), BOF (offer / retargeting)
- **On-creative copy** — any text baked into the asset (to avoid redundancy in ad copy)

Output a 4–6 line readout of what you saw. Keep going — don't pause for confirmation.

---

## Phase 3 — Scan the Account

Pull everything needed to propose without asking. Synthesize into 1–2 lines — never dump raw data at the user.

Collect:
- Top 1–3 campaigns by performance (last 30 days) — objective, optimization event, bid strategy, geos, age range, placements
- Top performing creatives — format, hook style, copy length (what's already resonating)
- Active pixel(s)
- Connected Instagram account(s)
- Existing custom audiences and lookalikes

**If no active campaigns exist:** scan paused campaigns and paused ad sets. Use the most recently paused or historically best-performing paused campaign as the structural reference — treat it identically to an active top performer. Note it's paused in the Source column ("Paused reference: `<campaign>`"). Only fall back to cold-start defaults if the account has zero campaigns of any status.

**Synthesis output (1–2 lines max):** "Ref: `<campaign>` — Sales / Purchase / broad US 25–55 / $80/day / Advantage+ / ROAS 3.2x [paused]. Pixel `<n>`, IG `<handle>`."

---

## Phase 4 — Propose or Ask

After Phases 2 and 3, the configuration table should be mostly filled. Default to proposing. Only ask if something is genuinely undetermined.

**Propose directly — no question — when:**
- Funnel stage matches an existing top campaign type
- A single pixel and IG account exist
- A reusable audience is available (CA, LAL, or broad that's performing)
- Budget magnitude can be inferred from top performers
- Optimization is obvious from the objective

**Ask only when:**
- Multiple ad accounts exist and the right one is unclear
- Creative funnel stage could legitimately be TOF or BOF
- Brand-new account with no reference data and budget is unknown
- Multiple conflicting pixels with no clear top-campaign preference

When asking is unavoidable — batch into one turn, maximum 2 questions, use tappable options not open text.

**Targeting search:** When building audience targeting for the ad set, use `facebook_search_targeting` to find interest/behavior/demographic IDs:
- `search_type: "adinterest"` + `query` — free-text search (e.g., "yoga", "anime", "running shoes")
- `search_type: "adinterestsuggestion"` + `interest_list` — expand from existing interests
- `search_type: "adTargetingCategory"` + `class` — browse by category (interests, behaviors, demographics, life_events, income, industries, family_statuses, user_device, user_os, etc.)
- Always use the returned `id` and `name` in `targeting.flexible_spec[].interests/behaviors/etc.`

---

## Phase 5 — Propose Configuration

Before launching the proposal, show a configuration summary table:

| Field | Proposed value | Source |
|---|---|---|
| Objective | Sales | Mirrors top campaign |
| Optimization event | Purchase | Account default |
| Daily budget | $50 | Median of top 3 campaigns |
| Bid strategy | Highest volume | Mirrors top campaign |
| Audience | Broad, US/CA, 25–55 | Top campaign + creative signals |
| Placements | Advantage+ | Mirrors top campaign |
| Pixel | `<n>` | Active pixel |
| Instagram account | `<handle>` | Active IG |
| Primary text | `<drafted from value prop>` | Creative analysis |
| Headline | `<drafted from hook>` | Creative analysis |
| CTA button | Shop Now | Matches creative CTA |
| Schedule | Continuous, starts now | Default |

Then send the proposal. The proposal itself handles user confirmation — don't add a "shall I proceed?" message after it.

---

## Phase 6 — Execute & Enable

Execution order is strict: campaign → ad set → ad+creative. Each step is propose → approve (auto-executes). Never merge steps.

> **Reminder:** Proposals auto-execute on approval. Capture IDs from the tool response after each approval. Do not call a separate execute tool.

> **Reminder — failure limit:** If any step below fails twice (zero items succeed), stop the entire workflow. See the HARD RULE section above.

### 6.1 — Create Campaign
-> **Read `meta-create-campaign` now.**
Follow its workflow exactly. Propose campaign -> user approves -> executes automatically -> capture `new_campaign_id` from the tool response.

### 6.2 — Create Ad Set
-> **Read `meta-create-adset` now.**
Follow its workflow exactly. Propose ad set (using `new_campaign_id`) -> user approves -> executes automatically -> capture `new_adset_id` from the tool response.

### 6.3 — Create Ad + Creative
-> **Read `meta-create-ad-with-creative` now.**
Follow its workflow exactly. Propose all ads in ONE batched call (using `new_adset_id`) -> user approves -> executes automatically -> capture all `new_ad_id` values from the tool response.

Campaigns and ad sets are created as **ACTIVE**. Ads are created as **PAUSED** — they are the user's final gate before spending begins. After all creations succeed, enable only the ads.

---

### Enable Sequence — Enable ads (1 call)

Campaigns and ad sets are already ACTIVE. Only ads need enabling. Batch ALL ads into ONE `facebook_propose_update_ads` call:

```json
facebook_propose_update_ads({
  "act_id": "act_XXX",
  "ads": [
    {
      "ad_id": "<new_ad_id_1>",
      "current_state": { "status": "PAUSED" },
      "new_state": { "status": "ACTIVE" },
      "description": "Enable ad: <ad_name_1>",
      "thumbnail_url": "<thumbnail_url_1>",
      "campaign_id": "<new_campaign_id>",
      "campaign_name": "<campaign_name>",
      "adset_id": "<new_adset_id>",
      "adset_name": "<adset_name>",
      "ad_name": "<ad_name_1>"
    },
    {
      "ad_id": "<new_ad_id_2>",
      "current_state": { "status": "PAUSED" },
      "new_state": { "status": "ACTIVE" },
      "description": "Enable ad: <ad_name_2>",
      "thumbnail_url": "<thumbnail_url_2>",
      "campaign_id": "<new_campaign_id>",
      "campaign_name": "<campaign_name>",
      "adset_id": "<new_adset_id>",
      "adset_name": "<adset_name>",
      "ad_name": "<ad_name_2>"
    }
  ]
})
```
-> user approves each ad individually -> each executes automatically on approval

**CRITICAL: 5 ads = 1 call with 5 items in the ads array. NOT 5 separate calls.**

**Never skip the enable approval step.** The user must explicitly approve going live — even if they said "just launch it" earlier.

---

## Edge Cases

- **No active campaigns, paused exist:** Use best paused campaign as reference. Label source as "Paused ref" in the table.
- **No campaigns at all:** Cold-start defaults — $20/day, broad targeting, Advantage+ placements, objective from CTA. Flag once, briefly.
- **Multiple creatives at once:** One campaign, one ad set, multiple ads. If funnel stages clearly differ -> separate ad sets.
- **"Just launch it":** Fast version — one reference lookup, then propose. Never skip the scan.
- **Cloning an existing ad:** Pull structure, swap creative, only surface what's changing in the table.
- **Off-brand creative:** One sentence flag in the readout. Still propose.
- **"Launch it live" / "Make it active":** Campaigns and ad sets are created ACTIVE. Ads are always created PAUSED first — the enable step is the user's final gate before spending begins.

---

## Output Rules

Every response the user sees should be short and scannable. No walls of text.

- **Phase 2 readout:** 2–3 lines max. Format / Hook / Funnel stage / CTA. No headers.
- **Phase 3 synthesis:** 1–2 lines. Campaign name, key numbers, pixel, IG. Done.
- **Phase 4 (if asking):** One short message framing the question + tappable options. No preamble.
- **Phase 5:** Configuration table + proposal. No follow-up "shall I proceed?" message.
- **Phase 6 enable:** No extra message after proposing the enable. The approval cards are the confirmation. After execution, one line: "Live. Ads are now active under `<campaign>`."
- Never show raw API output. Never repeat information already shown in a previous phase.

When to use

Coordinate Meta campaign, ad-set and creative creation with separate approval to enable new ads.

What you get

The complete original GoMarble guidance, with source attribution and declared limits.

How it works

Clarify scope, use available evidence, and apply the relevant source workflow within actual user authorization.

Requirements

Authorized account, business goal, creative assets, budget and live connector approval semantics.

Delivery and review notes

Real proposal approval may execute immediately; do not repeat execution based on the older operations skill. New ads stay paused until explicitly approved for activation; actual platform state must confirm the outcome. GoMarble connectors, referenced sibling skills and runtime dependencies are not bundled or installed by M11. Brief obvious-danger screening only; no functional test or comprehensive safety certification. Loading this text authorizes no external calls, account mutations or ad spend.

Starting prompt

Use Meta Ads Creation Workflow for [TASK]. Clarify Authorized account, business goal, creative assets, budget and live connector approval semantics. Separate observed evidence from heuristics and recommendations. Check actual connector availability and approval semantics before any external action.

Not for

Claiming installed GoMarble tools, proven campaign outcomes or authorization to launch ads merely from loading this source.

What M11 added

German routing and explicit scope. Original attribution: GoMarble. Real proposal approval may execute immediately; do not repeat execution based on the older operations skill. New ads stay paused until explicitly approved for activation; actual platform state must confirm the outcome. GoMarble connectors, referenced sibling skills and runtime dependencies are not bundled or installed by M11. Brief obvious-danger screening only; no functional test or comprehensive safety certification. Loading this text authorizes no external calls, account mutations or ad spend.

Original authorship remains with gomarble-ai/marketing-agent · Original source ↗

Original license & copyright

MIT License

Copyright (c) 2026 GoMarble

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

Source SHA-256: 8d1b084105d589e3da95a9654f0e1419460e68dd90e2e9fde0fca19eee16696f
Snapshot checked: 2026-09-28T16:19:52.867Z

Find a skill.Get to work.

All skills

Multi-Agent Orchestration

AI Context Window Audit

AI Agent Architecture Audit

AI Skill Discovery

Lead Magnet Strategy

Ideal Customer Profile

Go-to-Market Strategy

Marketing Campaign Ideas

Product Growth Loops

Competitor Analysis

North Star Metric

Product Positioning

Product Vision

Go-to-Market Motion Selection

Customer Feedback and JTBD Analysis

PESTLE Market Environment Analysis

Customer Journey Mapping

Ansoff Growth Options

Data Analysis Validation

Dataset Profiling

Statistical Analysis Guidance

Programmatic SEO Planning

Landing Page and Form Conversion Review

Paywall and Upgrade Planning

Signup and Registration Review

User Onboarding and Activation

Popup and Modal Planning

Email Sequence Copy and Flow

Marketing Campaign Planning

Marketing Content Drafting

Brand Voice and Content Review

Marketing Performance Reporting

Sales Company Research

Company and Contact Enrichment

Website Information Architecture

Content Strategy and Editorial Roadmap

Product Launch Planning

Customer Research and Voice of Customer

Community Growth and Member Experience

Competitor Research Profiles

Roadmap and Release Communication

Lifecycle Email Sequences

API Contracts and Interface Design

Observability and Instrumentation Planning

Agent Context and Session Handoff

Product Discovery Sprint

Deal Quality Scoring

Enrichment Waterfall Design

Customer Retention Playbook

Sales Coaching Practice

Revenue Cohort Analysis

Intent Signal Scoring

Customer Identity Matching

Segment Activation Planning

Sales Call Review

Retention Dashboard Design

Sales Coaching Competencies

Offer Delivery Design

Offer Value Communication

Service Productization

Conversion Hypothesis Prioritization

Braze Engagement Guidance

Brand Voice Guide Design

Content Pipeline Orchestration

Website Brand Profile

Buyer Objection Clarification

Klaviyo Integration Guidance

Offer Bonus Planning

Klaviyo Marketing Review

Cross-Platform Advertising Review

Expertise Business Model

LinkedIn Advertising Guidance

TikTok Advertising Guidance

Evidence Research Brief

Cinematic Video Prompts

Visual Prompt Model Adaptation

Visual Prompt Diagnostics

Short Film Development

Find a skill.
Get to work.