Skip to main content

Documentation Index

Fetch the complete documentation index at: https://adadvisor.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

The foundation skill is the floor every other AdAdvisor skill stands on. It loads automatically whenever you mention AdAdvisor, Meta (Facebook) ads, or your ad account, and it sets the conventions every chained skill builds on. You do not invoke it directly. Install it once and let it work in the background.

​
What does the foundation skill do?

Before any analysis or mutation, the foundation skill walks Claude through four mandatory checks:
1

Account discovery

Calls adadvisor:list_ad_accounts to find every account you have access to. If you have multiple, it asks which one to work on rather than guessing.
2

Data freshness check

Verifies data_synced is true on the chosen account. If the initial import is still running, Claude tells you to wait instead of working on incomplete numbers.
3

Business context grounding

Reads the adadvisor://account/{id}/context resource to pull your break-even ROAS, AOV, target CPL, daily budget cap, and brand details. Every recommendation downstream is scored against these.
4

Pixel sanity check

Confirms your Meta pixel is firing recently and matches your storefront URL. If it is broken, every performance read is unreliable. Claude flags this before going further.
After these four checks, Claude routes to the specific workflow skill (audit, diagnose, launch, etc.) based on what you asked for.

​
What mental models does the foundation skill load?

Beyond the four opening checks, the foundation skill loads four mental models that all other skills assume:

Unit economics

Break-even ROAS = 1 divided by contribution margin. Target ROAS = break-even multiplied by (1 + profit margin). Kill thresholds expressed in multiples of target CPA, never absolute dollars.

KPI decoding

Use result_count, cost_per_result, and conversion_result_name over legacy purchases and revenue. Subscribe and Lead campaigns return 0 on the legacy fields. Using them gives wrong answers.

Naming and currency

Currency values in major units always (50.0 means $50, not 5000 cents). Names auto-generate to a structured ADADVISOR || CBO/ABO || strategy || business || date pattern unless you have an existing convention.

Activation invariants

Every created entity starts PAUSED. Activation does not cascade. Resuming the campaign does not resume its ad sets or ads. You activate each level explicitly.

​
When does the foundation skill trigger?

The foundation skill loads on virtually any AdAdvisor session. Triggers include:
  • “adadvisor”
  • “meta ads”, “facebook ads”, “instagram ads”
  • “my ad account”, “look at my ads”
  • “let’s work on my account”, “connect my meta”
  • Any time another AdAdvisor MCP tool is about to be called
Other skills (audit, diagnose, launch, scale, creative, targeting, playbooks) all chain on top of the foundation. They assume the four opening checks have happened.

​
What anti-patterns does the foundation skill prevent?

Bad: “ROAS is 4.2x.” Good: “ROAS is 4.2x against your 3.3x break-even. Profitable, but margin is thinner than the 5x target you would usually scale on.”
Bad: passing daily_budget: 5000 and accidentally creating a $5,000 a day campaign. Good: always major units in account currency. The skill reminds Claude on every budget call.
Bad: “CPA is up today, let’s kill the campaign.” Good: “We’re at 1.4x target for one day, within normal variance. Kill threshold is 3x target with zero conversions over 3 days.”
Bad: Meta rejects with subcode 2490408. Good: the skill reminds Claude to pass the registrable second-level domain (e.g. example.com, not shop.example.com) on every pixel-optimized ad creation.
Bad: ads stay paused if their parent ad sets were not also resumed. Good: the skill instructs Claude to activate leaf-up (ad, then ad set, then campaign) and confirm state_after at each level.

​
How do I install the foundation skill?

Claude Code

The foundation ships with the full plugin. One install command, all eight skills.

Claude.ai & Desktop

Upload the foundation ZIP alongside any other skills you want.
Source code for this skill (including its references/) lives in github.com/AdAdvisor/skills. The whole repo is Apache 2.0. Fork it and modify the thresholds to match your business.

​
Frequently asked questions

The foundation skill performs four mandatory checks at the start of every session: account discovery (which Meta ad account to work on), data-freshness verification, business-context grounding (break-even ROAS, AOV, target CPL), and pixel sanity check. It also loads four mental models (unit economics, KPI decoding, naming/currency conventions, activation invariants) that every other AdAdvisor skill assumes.
Yes. Every other AdAdvisor skill (audit, diagnose, launch, scale, creative, targeting, playbooks) chains on top of the foundation. Without it, the other skills lack the break-even ROAS, target CPA, and pixel choice they need to score performance correctly.
Because Meta’s API uses minor units (cents) and the MCP server converts for you. Passing daily_budget: 5000 thinking in cents would create a 5,000adaycampaignratherthan5,000 a day campaign rather than 50. The foundation skill reminds Claude on every budget call to use major units in account currency.
Generic “good ROAS” is not actionable. The same 4x ROAS is profitable on a high-margin SaaS and unprofitable on a low-margin apparel brand. The context resource carries your break-even ROAS, AOV, target CPL, and daily budget cap so every recommendation is scored against your business, not industry averages.
On virtually any AdAdvisor session. Phrases like “adadvisor”, “meta ads”, “facebook ads”, “my ad account”, “look at my ads”, “connect my meta”, and any direct mention of an AdAdvisor MCP tool all load the foundation.

Audit

The next step after foundation completes. Pixel, then structure, then performance, then red flags.

Diagnose

Performance investigation grounded in the foundation’s unit economics.

Launch

Campaign creation that uses the foundation’s pixel choice and naming conventions.
Last modified on May 13, 2026