📦 Meta Ads Guide — 技能工具

Published by Didoo AI — Your AI Media Buyer. One that actually knows you, built for SMBs.
A skill system for Meta Ads strategy, execution, and analysis.

What is This System?

This is Didoo AI's Meta Ads Agent System — a set of 19 skills covering strategy, research, creative, execution, analysis, and reporting for Meta Ads campaigns.

For AI agents: This system is designed for modular use. Any skill can run independently or chain in sequence. Start with meta-ads-guide if you're unsure where to begin.

# Meta Ads Skill Routing Guide

When to Use

This is the entry-point navigator for Didoo AI's Meta Ads workflow. New here? Start at the diagram below. Not sure which skill to use? Jump to the Quick Routing Table.

---

Skill Relationship Map

REFERENCE / ENTRY
├── meta-ads-guide (this file)
│       ├── meta-ads-fundamentals (core mechanisms)
│       └── meta-ads-capi-setup (tracking setup)
PLAN -- NEW CAMPAIGN
├── meta-ads-strategy (campaign design)
│       └── meta-ads-research (competitive intel)
│       └── meta-ads-budget-planning (budget allocation)
EXECUTE -- LAUNCH & MANAGE
├── meta-ads-builder (creative generation)
│       └── [creative output]
└── meta-ads-publisher (create/update campaigns)
        └── meta-ads-scale-campaign (scaling)
        └── meta-ads-daily-pulse (daily monitoring)
ANALYZE -- INTERPRET & DIAGNOSE
├── meta-ads-daily-pulse (daily auto-scan)
├── meta-ads-healthcheck (on-demand check)
└── meta-ads-weekly-performance (weekly report)
        │
        ├── meta-ads-drop-diagnosis
        ├── meta-ads-analysis
        ├── meta-ads-lead-gen-analysis
        ├── meta-ads-creative-fatigue
        └── meta-ads-audience-analysis
                │
                └── meta-ads-recommendation (single exit point for all analysis)
                        │
                        └── meta-ads-publisher (execution)

Quick Routing Table

What you want to do	Use this skill
Entry point / not sure where to start	meta-ads-guide
"My ads are running -- are they healthy?"	meta-ads-healthcheck
"I want to understand why performance dropped suddenly"	meta-ads-drop-diagnosis
"I need to analyze campaign performance in depth"	meta-ads-analysis (general / e-commerce)
"I'm running lead generation campaigns and need analysis"	meta-ads-lead-gen-analysis (lead gen only — mutually exclusive with meta-ads-analysis; do NOT run both)
"I know what I want to do -- just do it (launch/change)"	meta-ads-publisher
"I have analysis results -- what should I actually do?"	meta-ads-recommendation
"I'm planning a new campaign -- where do I start?"	meta-ads-strategy
"I need ad creative -- copy, images, hooks"	meta-ads-builder
"I'm entering a new market -- what are competitors doing?"	meta-ads-research
"I have a winner and want to scale it"	meta-ads-scale-campaign
"My creative performance is declining over time"	meta-ads-creative-fatigue
"I want a structured weekly performance report"	meta-ads-weekly-performance
"I want daily monitoring signals before my morning review"	meta-ads-daily-pulse
"I need to understand how Meta's auction/bidding actually works"	meta-ads-fundamentals
"I need to set up CAPI / server-side tracking"	meta-ads-capi-setup
"I want to plan my ad budget across campaigns"	meta-ads-budget-planning
"I need to audit targeting, audience mix, or placement"	meta-ads-audience-analysis

---

Two Core Workflows

Workflow A -- New Campaign Launch

meta-ads-strategy -> meta-ads-research -> meta-ads-builder -> meta-ads-publisher
                                                                    |
                                                         meta-ads-daily-pulse
                                                                    |
                                                         meta-ads-analysis
                                                                    |
                                                         meta-ads-recommendation

Workflow B -- Problem Detection & Fix

meta-ads-healthcheck / meta-ads-daily-pulse
          | [flags problem]
meta-ads-drop-diagnosis
          | [root cause identified]
          ↓
    ┌─────────────────────────────────────────────┐
    │  Creative declining over time → creative-fatigue    │
    │  Audience or budget allocation → audience-analysis  │
    │  Lead gen campaign → lead-gen-analysis              │
    │  General / e-commerce → analysis                    │
    └─────────────────────────────────────────────┘
                              ↓
                   meta-ads-recommendation
                              ↓
meta-ads-publisher (execute) -> meta-ads-daily-pulse (monitor)

Important: meta-ads-analysis and meta-ads-lead-gen-analysis are mutually exclusive — they cover different campaign types. Choose one based on your campaign type. Do NOT run both in the same session.

---

Monitoring Cadence

Frequency	Skill	When
Daily -- 2 min	meta-ads-daily-pulse	Every morning
Weekly -- 10 min	meta-ads-weekly-performance	Every Monday
On-demand	meta-ads-healthcheck	When something feels off

---

Getting Started -- First Time Setup

Before running your first campaign:

• Connect your Meta Ads account -- Ad Account ID (act_XXXXXXXXX) from Ads Manager + Access Token from Meta Developer Console.
• Install Pixel + CAPI -- Without these, Meta can't optimize. See meta-ads-capi-setup.
• Define your first campaign goal -- Lead gen / Conversions / Link Clicks / Awareness.
• Set a realistic budget -- Minimum $10-15/day per adset. See meta-ads-budget-planning.
• Launch -- meta-ads-strategy -> meta-ads-builder -> meta-ads-publisher (always launch PAUSED, review, then activate).
• Wait 5-7 days -- Don't judge during Learning Phase. Changing things during Learning resets the clock.

---

Common First-Timer Mistakes

• Budget too low ($1-5/day) -- Learning Phase never completes
• Too many campaigns at once -- spread thin, none learns
• Expecting results in 24-48 hours -- give Meta time to optimize
• Changing too many things at once -- change one variable, then wait
• Not connecting CAPI -- CPA looks worse than reality

---

Key Constraint -- Analysis vs. Recommendation Boundary

• meta-ads-analysis (and its variants) -- explains what happened. Data only. Never tells you what to do.
• meta-ads-recommendation -- tells you what to do, why, and in what order. Requires analysis data first.
• Never output recommendations without running analysis first (or having existing data in session context).

---

Session Context Conventions

Analysis skills store their output in session context. Downstream skills read from there. This is the shared contract between skills.

Analysis -> Recommendation Data Flow

Skill (Writer)	Keys Written	Skill (Reader)
meta-ads-analysis	funnel_weak_points, trend_signals, anomalies, data_quality, lp_diagnosis_general	meta-ads-recommendation
meta-ads-lead-gen-analysis	lp_diagnosis, capi_status, cpl_breakdown, recommended_fix_priority	meta-ads-recommendation
meta-ads-audience-analysis	budget_reallocation_plan, audience_issues	meta-ads-recommendation
meta-ads-creative-fatigue	rotation_pipeline, fatigue_level, days_until_death	meta-ads-recommendation
meta-ads-drop-diagnosis	primary_root_cause, recovery_plan	meta-ads-recommendation

Cross-Skill Reference Rules

• Analysis skills never embed recommendation logic -- they store results and defer
• meta-ads-recommendation is the single exit point for all analysis outputs
• When one analysis skill routes to another (e.g., meta-ads-analysis -> meta-ads-lead-gen-analysis), preserve the original skill's context keys before overwriting
• When routing from meta-ads-recommendation back to execution, use meta-ads-publisher for all campaign changes

Session Key Collision — Resolved

meta-ads-analysis now writes lp_diagnosis_general; meta-ads-lead-gen-analysis writes lp_diagnosis. These keys no longer collide — they coexist. meta-ads-recommendation reads lp_diagnosis first (lead-gen-specific), falling back to lp_diagnosis_general if not present.