← Google Ads/Meta Ads LoomaScaleCONTENT HISTORY

Update to Google Ads/Meta Ads LoomaScale

Snapshot Nov 19, 2026 · 06:09 UTC · version 1.2.0

Collection source: downloaded plugin package.

WHAT CHANGED

First saved snapshot

No earlier snapshot is available to establish a change.

Save Google Ads/Meta Ads LoomaScale and check future changes in Following.

Up to 3 plugins free · no card required.

Compare saved observations

Download comparison JSON
Full technical diff · 0 changed fields
Full snapshot data
{
  "description": "Get started with LoomaScale or find useful next steps after a first audit. Use when the user asks what LoomaScale can do for their business, how to get value from connected marketing data, what to do next, or which other sources to connect. Tailor a first insight and repeatable daily or weekly workflows to their goals and work frustrations. Use the task-specific skills for an already defined report or change.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 618
    }
  ],
  "name": "getting-started",
  "skill_md_contents": "---\nname: getting-started\ndescription: Get started with LoomaScale or find useful next steps after a first audit. Use when the user asks what LoomaScale can do for their business, how to get value from connected marketing data, what to do next, or which other sources to connect. Tailor a first insight and repeatable daily or weekly workflows to their goals and work frustrations. Use the task-specific skills for an already defined report or change.\n---\n\n# Getting started with LoomaScale\n\nHelp the user turn advertising and marketing data into decisions that support their\nbusiness. Show what they can accomplish now, demonstrate one useful result, and\nleave them with a reason to return after today's issue is resolved. Personalize\nthe explanation around their work; avoid a feature tour or a comprehensive audit.\nOnboarding is read-only and does not authorize account or campaign changes.\n\n## Understand the business without an intake form\n\nReuse what the conversation already establishes: what the business sells, whom it\nserves, the user's role, the outcome they want, and work that is difficult or\nrepetitive. An agency may need client reporting; a shop may need more purchases;\na service business may need qualified enquiries. Treat these as possibilities,\nnot facts inferred from an account name.\n\nIf that context is missing, ask one short question combining the business and its\nmain frustration. For example: \"What do you sell, and what would help most right\nnow: more sales or qualified leads, less wasted ad spend, or less time reporting?\"\nUse the user's language and explain unfamiliar metrics. Do not repeat questions\nalready answered. While waiting, discover connections and give a brief explanation\nof what the available sources could help them decide. Keep any suggestions\nprovisional until their goal is known.\n\nFor a returning user, start with their previous finding or fix. Explain the next\nbusiness question it enables: whether the fix improved results, what else limits\ngrowth, or how to notice a recurrence. Skip introductory material they know.\n\n## Find what is connected and usable\n\nInspect the available tool schemas; client prefixes may differ. Call\n`loomascale_list_data_sources` with `{}` once, or reuse its recent result from this\nconversation. This free, read-only inventory returns the supported sources,\nconnection states and connection links. Use the returned catalog rather than a\nfixed list: availability can differ by user.\n\n- `connected` means a recorded provider login, not proof that a particular\n  account, property or container is accessible. `connected_logins_count` counts\n  logins, not accounts.\n- `not_connected` identifies a possible additional connection. Recommend it only\n  when it helps the user's goal and they use that service, or make that condition\n  explicit.\n- `needs_attention` and `unknown` need the returned explanation or a relevant\n  discovery read. Do not label either as disconnected or turn an error into zero.\n- `tools_in_current_plugin: false` means the current plugin cannot use that\n  source, even if connected. Explain the scope and use the returned `full_plugin`\n  installation links when relevant; reconnecting does not expand plugin scope.\n\nUse returned `connect_url` and `connections_url` values unchanged. If inventory is\nunavailable, use only relevant exposed discovery tools and explain that the full\nconnection state is unverified. An absent tool or empty account list does not\nprove a dashboard connection is missing. The fallback is the\n[LoomaScale connections dashboard](https://loomascale.com/dashboard/connections).\n\nSummarize the sources useful for this user's goal. Do not dump every account or\ncollect data from every connected platform just to demonstrate breadth.\n\n## Demonstrate one useful result\n\nChoose the smallest read that informs the user's main decision, normally from\none source or two clearly related sources. Discover only those resources using\nthe [integration workflow](../loomascale/SKILL.md). Pass explicit target IDs;\nnever use `*_select_account` during onboarding, because that changes a default\nshared across chats. If several businesses or client accounts are plausible,\nresolve the target before reading their performance. Do not choose the first or\nlowest-ID account arbitrarily.\n\nUse [cross-source-insights](../cross-source-insights/SKILL.md) for reporting dates,\ncomparisons, source matching and evidence rules. Unless the user supplies dates,\nuse the last 30 complete days against the preceding equal period; honor each\nsource's freshness and time zone. Start with a summary, then make a focused\ndrill-down only if needed. Stop when one decision has enough evidence; onboarding\ndoes not need an account-wide audit.\n\nExplain the finding in business terms, its source and dates, the next action to\nconsider, and how to measure whether that action helps. Keep observed facts,\nhypotheses and possible upside separate. If nothing material is wrong, show a\nuseful baseline or opportunity rather than manufacturing a problem. Sparse data,\nmissing business events, permission errors or plan limits are limitations, not\nevidence of poor performance. Follow the integration workflow's recovery guidance.\n\nKeep ad-platform attributed conversions separate from GA4 outcomes; neither\nproves profit or lead quality. Do not sum them into unique sales or promise growth,\nsavings or hours saved without evidence. Use known targets when available; do not\ninvent a target CPA, margin, budget or conversion event.\n\n## Make the ongoing value concrete\n\nSelect about three jobs that match the user's goal and available evidence,\nincluding something beyond fixing today's issue. Use this table as a menu, not\nas a checklist to display or execute in full.\n\nMake the path from insight to action clear. When relevant, explain that the user\ncan also ask LoomaScale to draft a campaign, research keywords, develop ad copy,\nor carry out supported Google Ads and Meta Ads campaign changes after review and\napproval. Mention supported analytics or tracking configuration changes only when\nthey help this user's problem. Show a practical loop: identify an opportunity,\nreview a proposed change, apply it when requested, then measure the same outcome.\nThe getting-started session itself remains read-only.\n\n| Daily work or business problem | Useful LoomaScale workflow | Why return? |\n|---|---|---|\n| Manually checking ads and spend | Review pacing, delivery and the largest spend changes in the active ad platform | A brief daily check identifies exceptions worth investigating. |\n| Not knowing why sales or enquiries changed | Compare relevant ad results with GA4 business events and landing-page or funnel changes | A weekly review helps choose the next improvement and check its effect. |\n| Paying for traffic that does not convert | Review search terms or creative performance, then relevant GA4 outcomes where linked | Revisit exclusions, creative tests and budget proposals as new evidence arrives. |\n| Visitors leaving without buying or enquiring | Examine GA4 landing pages, devices and ordered funnel steps | Prioritize a site experiment and compare the same outcome after it runs. |\n| Conflicting reports or unreliable tracking | Compare conversion definitions, GA4 and relevant GTM configuration; use Meta Pixel/Conversions API diagnostics for Meta | Recheck after tracking or site changes; configuration alone cannot prove live tag firing. |\n| Organic traffic or visibility declining | Use Search Console queries, pages and indexing, with GA4 outcomes when available | Review page and query opportunities and measure subsequent changes. |\n| Shopping products losing visibility | Review Merchant Center product eligibility and feed issues, plus related Google Ads evidence | Recheck product issues and whether affected products regain delivery. |\n| Running out of campaign or creative ideas | Research public Google and Meta ads using the competitor research skill | Develop a testable offer or message hypothesis, then evaluate the user's own results. |\n| Client reporting taking too long | Reuse a scoped performance brief for the selected client's sources, goals and dates | Repeat weekly with consistent definitions, results and next actions. |\n\nFor each selected job, provide a ready-to-send prompt tailored to the known\nbusiness, named target and metric, a useful cadence or trigger, and the decision\nthe result supports. Prefer \"Compare qualified-enquiry events for [known property]\nwith the previous week and explain the biggest change\" to \"Ask me for insights.\"\nUse only events actually identified; a lead event does not establish lead quality.\nMake unresolved names or choices explicit instead of inventing them.\n\nFor a business with little history or no campaigns, focus on supported preparation:\nkeyword opportunities when Google Ads is available, public competitor messaging,\na measurement baseline or a campaign proposal. Do not invent performance or\nrecommend launching paid campaigns without understanding the user's goal and\nbudget. For public ads, use\n[competitor-ad-research](../competitor-ad-research/SKILL.md); LoomaScale sign-in is\nrequired, but connecting the user's own ad account is not. Resolve the competitor\nand market before research. Public libraries reveal messaging, not competitor ROAS.\n\n## Recommend additional sources by the question they unlock\n\nSuggest at most one or two relevant sources from the returned catalog. Tie each\nto a current blind spot and ask whether the user already uses it if unknown:\n\n- Ads without GA4: connect an existing GA4 property to investigate what visitors\n  do after arriving, beyond the ad platform's attributed results.\n- GA4 without the relevant ad source: connect that ad platform to examine spend,\n  delivery and campaign context alongside site outcomes.\n- Organic acquisition: Search Console adds search queries, visibility and indexing;\n  GA4 adds on-site behavior and outcomes.\n- A shop with Shopping/feed problems: Merchant Center adds product eligibility\n  and feed diagnostics; use verified relationships before linking these to ads.\n- Tracking discrepancies: GTM adds configuration evidence when the site uses it;\n  Meta Ads includes available Pixel/dataset diagnostics without a separate source.\n\nThese are conditional examples, not reasons to connect everything. A useful\nexisting connection is enough to start. Do not present an irrelevant missing\nsource as a blocker. If a desired service is absent from the catalog, distinguish\nunsupported native connections from sources merely not connected. Do not promise\nCRM, store, payment, profit or lead-quality data that the catalog and exposed\ntools do not support. User-supplied exports can add context only when the current\nenvironment can actually read them.\n\nWith no usable connections, explain a relevant first workflow and link to its\nconnection path. Offer public competitor research if its tools and sign-in are\navailable; do not insist on an ad-account connection for that task. If tools or\naccess block a first result, state the exact next step and what can happen once\nit is resolved. Do not claim the analysis ran.\n\n## Leave a clear next step\n\nKeep the personalized brief compact, usually around 400–600 words after business\ncontext is established: how LoomaScale helps their goal, what the connected sources\nshow now, the selected repeatable jobs and prompts, and any useful extra connection.\nAdapt the length to the user's preference; do not make an unresolved clarification\nturn into a full report. When blocked on a connection or account choice, give the\nsmallest useful orientation and next step; defer the full set of jobs and prompts.\n\nFinish with one recommended next action and a specific return prompt: what to\ncheck, when or after which change, and which outcome to compare. After a fix,\ninclude checking its effect as well as the next improvement opportunity. Cadences\nare suggestions for the user to run; do not claim background monitoring, scheduled\nreports or notifications unless a supported host tool actually creates them after\nthe user requests it. Discovery does not authorize connections, writes or schedules.\nRoute later requested changes through the integration workflow and its approvals;\nLinkedIn Ads remains reporting-only.\n"
}

SHA-256 of public snapshot: 833740f4ac4f3c109ee311e028ca71fb91a2566cca392a3749e4c5ddf878d46c