/answer-ads-questions

name: answer-ads-questions
description: Answer questions about a brand's Meta ads (spend, results, which ad is winning, is it paused, why did Goose do that) only from reads made in the same turn, with the data window and sync time on every number and a plain caveat when the connection is stale or partial. Use it whenever a user asks how their ads are doing or why an ad action was taken.

Answer Ads Questions

**In one line:** answer "how are my ads doing?" from a tool you called this turn, say which days the numbers cover and when they were last synced, and say out loud when the data is incomplete.

• **Numbers come from tools, never from memory or the docs.** The `ads/` docs explain *why*

Goose did something. They are not a source of metrics.

• **Every number carries its window and sync time.** "You spent $212 from Sep 17 to Sep 23

(data last synced Sep 24, 09:00 UTC)."

• **Partial, stale or disconnected data is said straight after the answer**, never buried or left out.
• **This skill only reads.** It never pauses, changes or writes anything.

Where this skill's files are

Paths are relative to **this skill's own folder**, not your working directory (in GooseWorks: `agent-config/skills/answer-ads-questions/`). The harness docs contract ships with the `meta-ad-manager` skill, installed beside this one.

Choose one mode

Use the first mode that is available:

1. **GooseWorks** when the GooseWorks Meta read tools are callable. Read [references/gooseworks-adapter.md](references/gooseworks-adapter.md). 2. **Direct Meta** when there is no GooseWorks, but a Meta token with `ads_read` and the ad account id are available. Read [references/direct-meta-adapter.md](references/direct-meta-adapter.md). 3. **Planning only** when neither is available. Read [references/planning-only.md](references/planning-only.md).

**"The adapter"** below means the file for the mode you chose.

Purpose

Give a user a true, dated answer about their Meta ads that they could act on. The usual way this goes wrong is that the agent repeats a number from an earlier turn or a report, and presents it as current. Or it misses that half the account did not sync. This skill fixes the source (a tool read this turn) and the framing (window, sync time, connection state).

Reach for it on any question about spend, results, delivery, which ad is best or worst, whether a push is live or paused, what Goose has noticed, or why Goose made a decision.

Inputs

• **The user's question** (required), in their words.
• **The brand's `ads/` folder** (optional). Per the harness contract (the `meta-ad-manager` skill's `contract/RULES.md`):

`ads/README.md` first, then only the campaign in play. Read `ads/brand.md` for one field only: `familiarity`. A brand with no `ads/` folder has never run the harness. That is fine for metric questions. For "why" questions it means there is no recorded decision.

<!-- shared:answer-ads-questions start --> <!-- This block is maintained in ONE place and copied byte-for-byte into both the maintainers' source copy and the published goose-skills copy. Edit one, copy it to the other, and run the parity script before shipping either. -->

Workflow

1. Pick the read from the question

The adapter maps each read below to a real tool or API call, and says where each read's data window and sync time come from.

| The user asks… | Read | Also read | |---|---|---| | "How are my ads doing?", spend, results for a period | account totals for a window | the connection state, whenever it is not clean | | "Is Meta connected?", "why is data missing?" | the connection and sync state | the diagnostics, if the sync state does not explain it | | "Which campaign / ad set / ad is best or worst?" | every entity at that level. **Read every page** before you rank | — | | A trend, day by day, "since Tuesday" | a daily series for the entities in question | — | | One specific ad | that ad's context (performance, link to Goose, freshness). If the user gave a name, find the id from the ad list first | — | | "How are the ads Goose made doing?" | the performance of Goose's own published creatives | the connection state, for the sync time | | A Goose push: "did it go live?", "is it paused?" | that push's state, using the push id from the campaign's `state.md` (`meta_push_ids`) | — | | "Why did you pause / change / recommend that?" | `ads/README.md`, then that campaign's `decisions.md` | — | | "What have you noticed?", "anything wrong?" | the campaign's `state.md` `## Open recommendations` | fresh account totals |

**Only claim a read you made.** If the host has no read for something (alerts, a watch, a push record), never say "I checked" or "there are no alerts". Say what you did read, and that the rest is not visible from here.

2. Check the connection before you say any number

For any question that ends in a Meta number, read the connection state once in the turn, in addition to the numbers. Some reads report "ok" while the connection is only partly synced, so the numbers' own status is not enough. The adapter says which read carries the connection state and how each class below shows up.

| Class | What you say | |---|---| | **Clean**: connected, last sync finished | The numbers, with window and sync time. | | **Partial**: some parts of the last sync failed | **Read the sync detail first.** A caveat that only says "partial" or "some data may be incomplete" fails. Name the part that failed, in plain words ("ad images didn't finish syncing"). Say which numbers that part affects. Then give them. | | **Stale**: the last good sync is old | Give the numbers, say how old they are ("last synced 3 days ago"), and **make no recommendation**. | | **Not connected**: never connected, needs re-authorising, failed or disconnected | No numbers. Say Meta needs reconnecting (the adapter says where). Never ask for a token in chat. | | **Error**: the read failed, or there is no brand | Say you could n