Update to Adzviser Meta Ads

{
  "description": "Retrieve and analyze reporting from the user's connected Meta Ads accounts with Adzviser. Use for Facebook and Instagram advertising spend, clicks, purchases, campaign comparisons, trends, raw rows and Data Mode. Public competitor Ad Library searches and ad-account changes are not supported.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 250
    }
  ],
  "name": "adzviser-data-mode",
  "skill_md_contents": "---\nname: adzviser-data-mode\ndescription: Retrieve and analyze reporting from the user's connected Meta Ads accounts with Adzviser. Use for Facebook and Instagram advertising spend, clicks, purchases, campaign comparisons, trends, raw rows and Data Mode. Public competitor Ad Library searches and ad-account changes are not supported.\n---\n\n# Meta Ads Data Mode\n\nUse `retrieve_reporting_data` for questions about numbers. Use the preview skill\nonly for an explicitly requested interactive report. Do not open a widget merely\nto display a report already retrieved for analysis.\n\n## Reporting workflow\n\n1. Call `list_workspace` and select the user's named workspace. If access is\n   unavailable, explain the problem instead of substituting another workspace.\n   If no workspace was named and multiple suitable choices exist, ask which to use.\n2. Call `list_metrics_and_breakdowns_fb_ads` with the selected `workspace_name`.\n   Use exact returned labels, including workspace-specific custom conversions\n   when relevant. Request only the fields needed to answer the question.\n3. Resolve inclusive YYYY-MM-DD date pairs using the current date. Last month\n   means the previous complete calendar month; last week means the previous\n   complete Monday-Sunday week unless the user specifies otherwise. Dates use\n   the advertising account's reporting timezone.\n4. Send `adzviser_request` with `workspace_name`, `date_ranges`, optional\n   `time_granularity`, and `assorted_requests.fb_ads_request` containing `metrics`\n   and optional `breakdowns`. Include both periods in one comparison request.\n5. Read all tables, headers, rows and relevant `note_of_assumption` values under\n   `structuredContent.reportingData.FacebookAds`. The `data` object can contain\n   a table keyed by an empty string or by a breakdown such as PublisherPlatform.\n   Do not assume a fixed table key or flatten headers into data rows.\n6. Answer from the returned rows, with the workspace, exact dates, units and\n   relevant limitations. Reuse successful results instead of repeating queries.\n\nNever send conversation history, the raw user prompt, unrelated context,\ncredentials, email addresses or session tokens as report inputs. Use only\nstructured fields needed for the report. `reporting_mode=data` selects this\nworkflow; it is not a tool argument. Do not send `client_id`, `reporting_mode`\nor `_widget_action`. `set_reporting_mode` is an app-only preference control.\n\n## Interpret Meta Ads data\n\n- Include Account Currency for monetary reports. Preserve the reported currency\n  and keep different currencies separate. Do not assume USD or convert units.\n- Include Campaign ID with Campaign Name when distinguishing campaigns. Names\n  need not be unique. Use equivalent IDs for ad sets and ads when needed.\n- A request for conversions may mean purchases, leads or a custom event. Use\n  the event specified by the user; clarify an ambiguous conversion definition.\n  Website Purchases counts attributed purchase events, not unique customers.\n  Do not substitute Actions or a different conversion event for purchases.\n- Spend, Impressions, Link Clicks and Website Purchases are examples; confirm\n  current field labels through discovery. Link Clicks and Clicks (All) differ.\n- Publisher Platform can include Facebook, Instagram, Audience Network, Threads\n  or other returned values. Preserve actual groups rather than forcing every\n  row into Facebook or Instagram.\n- Calculate rates from their matching numerators and denominators, with N/A for\n  a zero denominator. Do not average percentages, frequencies or ROAS across\n  groups. Do not sum reach as though it were a deduplicated audience.\n- Use `time_granularity` for Date, Week, Month or Quarter instead of inventing\n  time breakdowns. Daily dates can be returned in a Date Range column.\n- Keep full returned precision for calculations and round only displayed money.\n  A zero comparison baseline has no finite percentage change. Missing rows are\n  not proof of zero activity; explain header-only or empty results honestly.\n- Configured ad-set targeting describes settings. Demographic performance\n  describes aggregate delivery; neither reveals individual audience members.\n\nOnly `fb_ads_request` is supported here. Do not use `meta_ad_library_request`\nfor connected-account reporting or claim this plugin can search public competitor\nads. It cannot edit budgets or bids, create or publish ads, pause campaigns,\ndelete records, or disclose individual people who saw an ad.\n\nDo not invent arbitrary filters, URLs, sorting, limits or query-text inputs.\nFor unsupported fields, refresh discovery, correct the request once and retry.\nExplain authentication or access errors without claiming the report succeeded.\n"
}