A portable, analysis-first skill for Facebook and Instagram advertising.
Meta Ads Intelligence helps an AI agent:
- analyze Meta Ads performance
- diagnose campaign, ad set, and ad problems
- compare creatives
- detect fatigue and anomalies
- recommend budget/scaling decisions
- analyze public competitor ads
- break down hooks, offers, messaging, and CTAs
- generate decision-oriented reports
- perform account actions when the user explicitly asks and the environment has the required Meta API/tool access
Analysis is the default. Actions require explicit intent.
For example:
Analyze my campaign
stays read-only.
But:
Pause this campaign
can enter action mode, subject to access and confirmation rules.
Meta-Ads-Intelligence/
├── SKILL.md
├── README.md
├── references/
│ ├── meta-marketing-api.md
│ ├── campaign-diagnosis.md
│ ├── competitor-analysis.md
│ ├── creative-analysis.md
│ ├── reporting.md
│ └── safety-and-confirmation.md
├── scripts/
├── schemas/
├── examples/
└── evals/
The skill can work with:
- live Meta Marketing API/connector data when actually connected
- Ads Manager exports
- screenshots and reports supplied by the user
- public competitor/ad-library information
It must never pretend to have live access when no connection exists.
Public competitor research is observational. The skill must not present a competitor's private spend, CPA, ROAS, revenue, exact targeting, or conversion results as known facts unless reliable evidence is supplied.
Consequential changes such as publishing, deleting, budget increases, pausing, targeting changes, and creative replacement are handled through an explicit proposal/confirmation workflow.
The core skill is platform-agnostic. The repository separates reasoning instructions from API/tool adapters so it can be adapted to Claude Code, agentic IDEs, and other compatible skill systems without embedding credentials or vendor-specific secrets.
Early development — core routing, analysis rules, competitor intelligence, reporting, API guidance, and action-safety rules are implemented. Tool adapters, reusable analysis scripts, schemas, examples, and evaluation cases are added incrementally.