agentleFS
Sign inStart free
gomarble-aigomarble-ai/marketing-agent/skills/diagnose-performance/SKILL.md

diagnose-performance

Skill15 starsChanged 37 days agoMITView on GitHub

Use when the user asks why performance changed or what's going on with an account: ROAS dropped, CPA spiked, spend jumped, conversions fell, blended results slipped, or 'what moved this week'. Traces the move across Meta, Google, TikTok, LinkedIn and Microsoft Ads plus GA4 and Shopify to the real driver, then proposes the smallest fix. Also for account audits and weekly performance reads.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Call recall_memory for the account: its baselines, targets, known patterns and earlier findings.

    line 20
  2. Distinctive:

    Check that the change is outside normal variance and isn't partial data.

    line 29
  3. Distinctive:

    Check volume. Ten conversions moving to eight is noise, not a trend.

    line 30
  4. Very specific:

    Before any recommendation, read meta-guardrails or google-ads-guardrails.

    line 51
  5. Very specific:

    If the user wants to apply a fix, hand off to the matching skill: shift-budget, clean-wasted-spend or manage-campaigns.

    line 69

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Diagnose performance

Find the cause before changing the account. Trace a performance move to its real driver, then recommend the smallest action that fixes it.

When to use

  • "Why did ROAS drop last week?" / "Why is CPA up?"
  • "What changed in my account?" / "What moved this week?"
  • "Audit my Meta (or Google) account."
  • Blended results slipped and the user doesn't know which channel caused it.

Before you start

  1. Find the accounts (see get-started) and confirm the currency.
  2. Call recall_memory for the account: its baselines, targets, known patterns and earlier findings. Judge the move against this account's normal range, not a universal benchmark.
  3. Settle the comparison: the period in question vs the one before it (for example last 7 days vs the prior 7), and the target KPI (ROAS, CPA, CPL). If there's no target, use the account's trailing 30-day average and say so.

Workflow

1. The movement

Confirm the move is real before explaining it.

  • Pull both periods at account level for every channel involved.
  • Check that the change is outside normal variance and isn't partial data. Recent days can still be missing conversions.
  • Check volume. Ten conversions moving to eight is noise, not a trend.

2. The driver

Drill down one level at a time until the change is isolated to a specific line.

Where to lookWhat it can show
ChannelWhich channel owns the move. Compare with GA4 sessions and conversions by channel, and Shopify revenue, to separate platform attribution from real business change.
Campaign → ad set / ad group → adWhere spend and results actually shifted. Quantify it: "Campaign X is 60% of spend and 90% of the CPA increase."
Changes in the accountBudget, bid, targeting, status or creative edits right before the move.
CreativeHook rate, hold rate and CTR decay together with rising frequency means fatigue (see analyse-creative).
Audience and placementDelivery shifting to a worse segment, placement or geography.
Search termsNew low-intent queries or match-type leakage (see clean-wasted-spend).
AuctionCPM or CPC up with impression share down means competition, not quality.
TrackingConversions falling while GA4 or Shopify stay flat usually means a tracking or attribution break, not a performance problem.

Apply the platform methodology:

  • Meta: meta-performance-analysis, with meta-depth-of-analysis. Read meta-custom-event-interpretation before judging any sales campaign that optimizes for COMPLETE_REGISTRATION.
  • Google: google-ads-search-analysis, google-ads-shopping or google-ads-pmax-evaluation, with google-ads-depth-of-analysis.
  • TikTok, LinkedIn, Microsoft Ads: the matching channel skill.
  • Before any recommendation, read meta-guardrails or google-ads-guardrails.

3. The proposed actions

Recommend the smallest change that addresses the driver. Refreshing creative is often better than cutting budget. For each action give the entity, what to change, the evidence, and the expected effect. Then surface one or two things the user didn't ask about but should know: a trend that will become a problem in the next 7–14 days.

Output

Use three parts:

  1. The movement. The metric, the change and the period, against the account's normal range. Example: "ROAS 2.1 vs a normal 2.8–3.2, down 27% week over week."
  2. The driver. The root cause with evidence. Example: "Hook rate and conversion rate fell together on the two top-spend ads, which are 70% of spend."
  3. The proposed actions. A ranked list with entity, change, reason and expected effect.

End by offering to make the changes (with approval) or to keep watching with an agent.

Acting on it

If the user wants to apply a fix, hand off to the matching skill: shift-budget, clean-wasted-spend or manage-campaigns. Every change goes dry run → the user approves in chat → live (see get-started).

Put it on a schedule

Offer a daily "performance analyst" agent that flags the next meaningful move (see automate-with-agents).

Tools

SourceTools
Meta Adsfacebook_get_details_of_ad_account, facebook_get_adaccount_insights, facebook_get_async_adaccount_insights (large pulls), facebook_fetch_pagination_url, facebook_list_campaigns, facebook_list_adsets, facebook_list_ads, facebook_get_campaign_details, facebook_get_adset_details, facebook_get_activities_by_adaccount (change history)
Google Adsgoogle_ads_run_gaql, google_ads_get_change_logs (changes joined with performance)
TikTok Adstiktok_get_basic_report_enhanced, tiktok_get_async_report, tiktok_get_audience_report, tiktok_get_campaigns, tiktok_get_adgroups, tiktok_get_ads, tiktok_fetch_pagination, tiktok_get_change_log
LinkedIn Adslinkedin_get_ad_analytics, linkedin_get_campaigns
Microsoft Adsbing_ads_get_performance, bing_ads_get_change_history
GA4google_analytics_run_report, google_analytics_get_traffic_sources (read ga4-source-of-truth first)
Shopifyshopify_run_analytics_query (read shopify-order-discipline first)
Memoryrecall_memory

Reach, frequency and rate metrics can't be summed across rows. Recompute them from totals.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free