Menu
Use when launching a Google Ads experiment. Loaded after the create master skill.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Use when researching competitor ad creatives: find competitors, brand IDs, evergreen winners, breakout winners, pattern analysis.
Use when generating Word documents (.docx). Uses the marble_docx API.
Use when generating PDF documents. Uses the marble_pdf API.
Use when generating PowerPoint presentations (.pptx). Uses the marble_pptx API.
Use when generating Excel workbooks (.xlsx). Pandas + openpyxl patterns for multiple sheets, number formats, freeze panes, formulas.
Use for any GA4 query whose result will be reported back to the user. Covers conversions ≠ transactions distinction, channel-subset-sum trap with sessionDefaultChannelGroup, intraday partial-data discipline, Paid Social aggregation rules, API-vs-UI attribution mismatch.
| name | google-ads-create-experiment |
| description | Use when launching a Google Ads experiment. Loaded after the create master skill. |
Tools: google_ads_propose_create_experiment, google_ads_propose_update_experiment. Tool schemas describe shape; this skill covers the business rules.
type: SEARCH_CUSTOM only. No DISPLAY_CUSTOM, no SHOPPING, no YOUTUBE. Reject any other type up front.base_campaign_id: must reference an existing campaign with advertising_channel_type = SEARCH.Two-arm A/B test on a single base campaign:
SETUP), then both ExperimentArms — control reuses base campaign; treatment forks a draft campaign.draft_modifications apply to the treatment's draft campaign.experiments/{id}:scheduleExperiment (Google long-running operation) and polls until done.traffic_split.control + traffic_split.treatment must sum to 100. Each in 1–99.start_date: ≥ today + 1 day, within ~1 year.end_date: > start_date.name is unique per customer — duplicate names reject. Append a suffix and retry.draft_modifications.daily_budget is propose-side advertised but the executor does NOT push it. Tell the user budget changes on the draft campaign aren't supported in V1; propose a separate post-schedule update if they want one.We expose ENABLE/PAUSE-style transitions and Google's lifecycle promotions. We do not expose entity removal.
action | Effect | Allowed from |
|---|---|---|
RENAME | Rename experiment | any state |
PAUSE | Pause running experiment | RUNNING |
RESUME | Resume paused experiment | HALTED |
END | End experiment immediately (LRO) | RUNNING / GRADUATED |
PROMOTE | Promote treatment to base (LRO) | RUNNING / ENDED |
GRADUATE | Convert treatment into a standalone campaign (LRO) | RUNNING |
END, PROMOTE, GRADUATE use Google's long-running operation system. The executor polls until done; failures show up in the LRO's error field. Common LRO errors:
START_DATE_TOO_FAR_IN_FUTURECANNOT_CREATE_EXPERIMENT_CAMPAIGN_WITH_SHARED_BUDGETSurface these directly to the user — they're fixable in the Ads UI by adjusting dates or detaching the shared budget.
Create:
Done. Created and scheduled experiment
<name>(id<id>).<control>/<treatment>split on campaign<base_name>. Treatment draft:<draft_name>. Will start<start_date>.
Update:
Done.
<action>on experiment<name>(id<id>).