Tools are the bricks. Skills are the instructions.
The media buyer's job,written for an agent.
200 skills: what to run, in which order, on which tools, and the numbers at which a conclusion stops being safe. Each one states where it refuses to conclude and where a human takes over.
- 200
- Skills
- 26
- Deep contracts
- 174
- Recipes
- 167
- Tools cited
The deep contracts.
Written by hand, with quantified decision rules, the evidence every conclusion has to cite, and the thresholds below which the agent must refuse to conclude at all.
26 of 200
Account audit
Turn an unfamiliar ad account into a ranked list of problems and confirmations, each backed by numbers.
Weekly performance review
Produce the week's cross-platform picture with explanations, not just deltas.
Creative fatigue
Separate creatives that are worn out from creatives that never worked, and rank refresh candidates by spend at risk.
Search waste
Produce the exact list of terms to negative, with the spend each one burned as the proof.
PMax diagnostics
Make an opaque campaign type inspectable: placements, asset health and spend, in one ranked view.
Cross-platform performance
Answer 'which platform earns its budget' with comparable numbers and explicit caveats, not four self-graded report cards.
Budget reallocation
Reallocate with discipline: right sizes, right order, previews first, and a rollback path.
Tracking audit
Establish whether measurement is sound before anyone optimises against it.
Anomaly investigation
Name the cause of a sudden change, with evidence, in minutes, checking causes in the order of their base rates.
Agency client review
Assemble a client-ready narrative where every figure survives the client's own dashboard check.
Meta delivery diagnosis
Name the binding constraint on a Meta ad set that will not spend, using Meta's own diagnostics before any theory.
Shopping feed health
Find the products the platforms refuse to serve, and rank feed fixes by the revenue they unlock.
Product performance
Rank products by advertising efficiency with names, not item IDs, so the merchandising decision is readable.
Keyword expansion
Deliver a keyword plan the account can actually absorb: net-new terms, sized, grouped, costed.
Paid and organic overlap
Quantify the overlap between paid and organic search, and separate defensible spend from probable substitution.
Organic decay triage
Separate the three causes of organic loss that get confused, because each has a different owner and fix.
Indexation health
Keep the revenue pages indexed and freshly crawled, and catch coverage regressions while they are still cheap.
Audience overlap
Stop paying auction premiums against yourself, with the overlapping pairs named and sized.
Lift study readout
Turn a lift study into a decision-grade readout: effect, confidence, caveats, and the decisions it does and does not support.
Spark Ads review
Decide where Spark Ads earn their place in the mix, on numbers instead of platform folklore.
Pinterest planning
Plan Pinterest launches on the platform's demand data instead of transplanting Meta assumptions.
Funnel analysis
Name the step where users leave and the segment where it is worst, with measurement verified before conclusions.
Landing page audit
Rank paid landing pages by the money they lose, with the cause class attached: broken, slow, or unconvincing.
New client onboarding
Compress the first week of account discovery into one structured pass with a written baseline.
Campaign launch QA
Catch the launch errors that cost real money in the first 48 hours, before the first impression.
Bid strategy review
Judge bid strategies on the platform's own counterfactuals instead of folklore, and find the strategies working against their budgets.
The other tier
174 recipes: shorter contracts on the same real tools, each with at least one decision rule and one trap.
Every skill, grouped by what it needs.
A skill sits under the source it cannot work without. The last group needs no single platform: those are the ones that read several and reconcile them.
200 skills
200 skills
- Meta Ads31
- Pinterest Ads25
- Google Ads54
- TikTok Ads23
- Search Console20
- Google Analytics 429
- Cross-platform18
Meta Ads
31 skills4 deep- Deep contract: Creative fatigueSeparate creatives that are worn out from creatives that never worked, and rank refresh candidates by spend at risk.
- Deep contract: Meta delivery diagnosisName the binding constraint on a Meta ad set that will not spend, using Meta's own diagnostics before any theory.
- Deep contract: Audience overlapStop paying auction premiums against yourself, with the overlapping pairs named and sized.
- Deep contract: Lift study readoutTurn a lift study into a decision-grade readout: effect, confidence, caveats, and the decisions it does and does not support.
- Campaign readoutCampaign performance ranked by spend, with objectives and buying types visible.
- Ad set readoutThe ad set table for one campaign: audiences, budgets, delivery, results.
- Ad readoutAd-level results joined to their creatives, so findings name assets.
- Creative inventoryWhat creative actually runs: formats, ages, reuse across campaigns.
- Frequency checkFrequency by ad set against tolerance, with reach context.
- Audience inventoryEvery audience, its definition, size band and the ad sets using it.
- Pixel healthWhether the pixel and its events still feed the account truth.
- Page posts reviewWhich page posts are being amplified and what they carry.
- Instagram assets checkWhich Instagram accounts are wired, and what placement stories rely on them.
- Catalogue checkWhether the catalogue serves what the ads assume.
- Structure mapThe campaign tree with objectives, budgets and buying types, on one page.
- Activity log reviewWho changed what on the account, aligned against delivery moves.
- Brand safety reviewWhat the account excludes and allows, against brand policy.
- Business assets mapWhich pages, pixels, catalogues and accounts exist, and what the connection reaches.
- Token health checkWhether the connection itself is about to fail, before it does.
- Entity searchResolve names to IDs across campaigns, ad sets and ads.
- Experiments inventoryThe account's experiment history: what was tested, what concluded.
- Organic contextWhat the brand's organic presence adds to or explains about paid results.
- Spend drop checkWhy Meta spend fell: edits, delivery constraints, or auction.
- CPA spike checkWhy Meta CPA jumped: pixel, fatigue, auction, or edits.
- Naming fix runRename campaigns, ad sets and ads to the convention, previewed each.
- Monthly packThe Meta month in one document: results, movers, work done, next month.
- Budget pacingWhether Meta spend lands the month on target, with the platform's own budget mechanics respected.
- Placement readoutFeed, Stories, Reels, Audience Network: who earns, who leaks.
- Country readoutMarket-level truth inside multi-country campaigns.
- Video metrics readoutThruPlay, holds and drop-offs: what the video does before the click.
- Budget mode mapWhere the budget lever actually sits, campaign by campaign.
Pinterest Ads
25 skills1 deep- Deep contract: Pinterest planningPlan Pinterest launches on the platform's demand data instead of transplanting Meta assumptions.
- Campaign readoutCampaign performance ranked by spend, objectives visible, archived counted.
- Delivery metrics readoutImpressions, clicks, saves and spend, in Pinterest's own grammar.
- Pin analyticsWhich pins carry the account, organic and paid sides both.
- Audience insights readoutWho engages the account: interests, categories, the planning raw material.
- Audiences inventoryThe audience estate mapped to ad groups.
- Keyword intelligence readoutSearch terms and volumes in Pinterest's own search, for targeting and creative language.
- Trends readoutWhat rises, when it peaks, for the market in question.
- Targeting mapWhat the plan's segments translate to in Pinterest's taxonomy.
- Catalogue diagnosticsFeed ingestion and product status on the Pinterest side.
- Catalogue performanceWhich products Pinterest actually sells, at SKU grain.
- Conversion setup checkWhether the tag and events feed the account truthfully.
- Billing reviewWhat was billed, against what was planned.
- Organic inventoryWhat organic presence exists to amplify or learn from.
- Lead assets checkWhether lead forms and their wiring stand ready.
- Delivery estimateWhat the drafted targeting buys at the intended budget.
- Specialized exportsThe exports packaged reports do not cover.
- Targeting reportWhich targeting choices earn: interests, keywords, demographics.
- Business assets mapWhat exists and what the connection reaches.
- Platform resourcesThe platform-level reference data behind the account reads.
- Spend drop checkWhy Pinterest spend fell: statuses, budgets, or delivery.
- CPA spike checkWhy Pinterest CPA jumped: tag, lag, or auction.
- Monthly packThe Pinterest month in one document: results, movers, work done, next month.
- Budget pacingWhether Pinterest spend lands the month on target, with the platform's own budget mechanics respected.
- Country readoutMarket-level truth for multi-market Pinterest accounts.
Google Ads
54 skills16 deep- Deep contract: Account auditTurn an unfamiliar ad account into a ranked list of problems and confirmations, each backed by numbers.
- Deep contract: Weekly performance reviewProduce the week's cross-platform picture with explanations, not just deltas.
- Deep contract: Search wasteProduce the exact list of terms to negative, with the spend each one burned as the proof.
- Deep contract: PMax diagnosticsMake an opaque campaign type inspectable: placements, asset health and spend, in one ranked view.
- Deep contract: Cross-platform performanceAnswer 'which platform earns its budget' with comparable numbers and explicit caveats, not four self-graded report cards.
- Deep contract: Budget reallocationReallocate with discipline: right sizes, right order, previews first, and a rollback path.
- Deep contract: Anomaly investigationName the cause of a sudden change, with evidence, in minutes, checking causes in the order of their base rates.
- Deep contract: Agency client reviewAssemble a client-ready narrative where every figure survives the client's own dashboard check.
- Deep contract: Shopping feed healthFind the products the platforms refuse to serve, and rank feed fixes by the revenue they unlock.
- Deep contract: Product performanceRank products by advertising efficiency with names, not item IDs, so the merchandising decision is readable.
- Deep contract: Keyword expansionDeliver a keyword plan the account can actually absorb: net-new terms, sized, grouped, costed.
- Deep contract: Paid and organic overlapQuantify the overlap between paid and organic search, and separate defensible spend from probable substitution.
- Deep contract: Landing page auditRank paid landing pages by the money they lose, with the cause class attached: broken, slow, or unconvincing.
- Deep contract: New client onboardingCompress the first week of account discovery into one structured pass with a written baseline.
- Deep contract: Campaign launch QACatch the launch errors that cost real money in the first 48 hours, before the first impression.
- Deep contract: Bid strategy reviewJudge bid strategies on the platform's own counterfactuals instead of folklore, and find the strategies working against their budgets.
- Campaign readoutA clean campaign-level performance table, ranked by spend, with statuses attached.
- Ad group readoutThe ad-group table for one campaign, spend-ranked, with statuses.
- Keyword performanceKeywords ranked by spend with quality signals, cut to what is judgeable.
- Search terms miningThe real queries behind the keywords: converters to add, junk to negative.
- Budget pacingWhether spend will land on target, campaign by campaign, with shared budgets untangled.
- Change history reviewWho changed what, when, on what — the account's black box recorder.
- Conversion actions reviewWhich conversion actions count, which bid, which are dead, and what 'conversions' actually sums.
- Recommendations triageGoogle's suggestions sorted into worth-it, ignore, and actively-harmful.
- Geo performanceWhere the money works: performance by location against the targeting's intent.
- Device performanceWhether mobile, desktop and tablet earn their shares, campaign by campaign.
- Ad schedule reviewWhether the schedule matches when conversions actually happen.
- Shopping readoutProduct-level Shopping truth: what serves, what sells, what silently sits.
- PMax placements readoutThe placement distribution PMax does not volunteer.
- PMax asset coverageWhich asset groups lack the assets Google says it wants.
- Simulations readoutWhat Google's own counterfactuals say a change would trade.
- Account mapThe MCC tree with names, so nobody works on IDs.
- Custom GAQL reportThe exact table nothing pre-built provides, validated before it runs.
- Keyword ideasFast candidate terms with volumes, for a brief or a brainstorm.
- Keyword forecastWhat a keyword list would buy at a bid level, from Google's forecast.
- Geo target lookupThe exact geo target constants for the market brief, ambiguities resolved.
- Paid and organic quick checkThe combined view for specific queries, without the full overlap study.
- Bidding strategy mapThe strategy inventory: who optimises what, toward which target.
- Landing pages readoutWhich URLs take the clicks, and their mobile health, in one table.
- Brand vs non-brand splitThe two economies of a search account, separated before anyone quotes an average.
- Impression share readoutHow much of the available auction the account wins, and whether budget or rank loses the rest.
- Disapproval scanWhat policy stopped, before it is discovered as silent underdelivery.
- Spend drop checkWhy Google spend fell: humans, budgets, policy or auction, in that order.
- CPA spike checkWhy Google CPA jumped: measurement, mix, competition or bids.
- No-delivery checkWhy a campaign shows zero impressions: the eligibility chain, walked.
- Naming fix runRename the violators the naming audit found, previewed one by one.
- Monthly packThe Google Ads month in one document: results, movers, work done, next month.
- Ad copy reviewWhich headlines and descriptions carry the ads, from the asset view.
- Quality Score reviewWhere QS is lost: expected CTR, relevance, or landing experience.
- PMax vs Search overlapWhether PMax buys queries Search already wins, and at what cost difference.
- Audience readoutWhich audience segments earn across campaigns.
- Assets and extensions readoutWhether the extensions earn their slots.
- Negative conflict checkWhere the negative lists fight the account's own winners.
- Video campaigns readoutVideo campaigns judged on video economics, not search metrics.
TikTok Ads
23 skills1 deep- Deep contract: Spark Ads reviewDecide where Spark Ads earn their place in the mix, on numbers instead of platform folklore.
- Campaign readoutCampaign performance ranked by spend, objectives visible.
- Ad group readoutAd groups with targeting, placements and results for one campaign.
- Ads readoutAd-level truth including the archived inventory most reads silently drop.
- Video assets reviewWhich videos run, their ages, and reuse across ad groups.
- Delivery checkWhat is actually serving, what is rejected, what is limited, right now.
- Audience inventoryThe audience estate with sizes and the ad groups using each.
- Events checkWhether the events the account optimises to actually fire.
- Spark inventoryWhich organic posts run as ads, and which authorisations expire when.
- Search readinessWhether TikTok search is worth entering now, on the account's own signals.
- Targeting mapWhat targeting exists for the plan's segments, from the catalogue itself.
- Shop catalogue checkWhether the shop catalogue serves what commerce campaigns assume.
- Raw reportThe exact table the packaged reads do not provide.
- Advertiser info checkCurrency, timezone and status per advertiser, the merge prerequisites.
- Entities raw probeDirect entity access for fields the packaged tools do not surface.
- Read endpoint probeA GET passthrough for reads the catalogue has not packaged yet.
- Spend drop checkWhy TikTok spend fell: delivery status, review, budgets, or auction.
- CPA spike checkWhy TikTok CPA jumped: events, fatigue, or auction seasonality.
- Monthly packThe TikTok month in one document: results, movers, work done, next month.
- Budget pacingWhether TikTok spend lands the month on target, with the platform's own budget mechanics respected.
- Placement readoutWhere TikTok delivery lands and what each surface earns.
- Country readoutMarket truth inside consolidated advertisers.
- Video metrics readoutWatch time, completions and drop points per creative.
Search Console
20 skills2 deep- Deep contract: Organic decay triageSeparate the three causes of organic loss that get confused, because each has a different owner and fix.
- Deep contract: Indexation healthKeep the revenue pages indexed and freshly crawled, and catch coverage regressions while they are still cheap.
- Search analytics readoutClicks, impressions, CTR and position, with freshness checked first.
- Losses and gainsWhat moved, ranked by click impact, in one pass.
- Cannibalisation checkWhether the site's own pages compete on the queries that matter.
- Query clusteringThe site's queries grouped into themes a strategy can address.
- Search appearance trendsWhether rich results and features move CTR independent of rank.
- Search types compareWhich search surfaces carry the site, and which are untapped.
- URL inspectionGoogle's own verdict on one page: indexed, canonical, last crawl.
- Bulk inspectionIndex verdicts across a template or section, quota-aware.
- Sitemap inventoryWhat sitemaps exist, when processed, with what result.
- Sitemap healthErrors, warnings and coverage per sitemap file.
- Sitemap deltasThe site's own publication log, as Google sees it.
- Freshness watchWhether Google re-reads the pages whose content moves.
- Large-site sampling planReliable index conclusions on a million-page site without a million inspections.
- Site inventoryWhich properties exist and which are ticked.
- Page readoutWhich pages carry organic, the page-side of the query view.
- Country readoutOrganic market map: where search brings whom.
- Device readoutOrganic device split: positions and CTR per device class.
- Brand splitThe two organic economies, separated before any trend claim.
Google Analytics 4
29 skills2 deep- Deep contract: Tracking auditEstablish whether measurement is sound before anyone optimises against it.
- Deep contract: Funnel analysisName the step where users leave and the segment where it is worst, with measurement verified before conclusions.
- Traffic readoutSessions, users and engagement over the window, with the channel cut.
- Channel performanceEach channel's sessions, key events and revenue, the site-side view of media.
- Key events reviewWhich events count as conversions, and whether they fire plausibly.
- Event parameters checkDeclared parameters versus filled ones, the metadata truth.
- Ecommerce readoutRevenue, items and funnels from the site's own till.
- Funnel quick readThe drop-off curve for an already-instrumented funnel.
- Realtime checkWhether an event fires right now, during a launch or a fix.
- Pivot analysisThe two-way cut a flat report cannot show.
- Audience inventoryWhich audiences exist, their sizes, their diagnostics.
- Audience export runThe audience's actual membership, exported and quantified.
- BigQuery export checkWhether the raw-data pipeline downstream teams rely on is intact.
- Server-side tagging checkWhether the server container still relays truthfully.
- Quotas checkHow much API budget remains before a big pull.
- Property config reviewTimezone, currency, retention, the settings every number inherits.
- Compatibility checkWhether the planned report's fields can legally combine.
- Batch reportingSeveral cuts in one flight, quota-shaped.
- Property inventoryWhat the connection can read, mapped and ticked-aware.
- Metadata lookupWhat fields exist on this property, custom definitions included.
- Query validationCatch the shape errors before they cost a quota token.
- Signup funnelThe registration journey's drop-off, step by step.
- Checkout funnelCart to purchase, the retail money path.
- Lead funnelVisit to qualified lead, the B2B money path.
- Landing pages readoutWhich entry pages earn engagement and conversions.
- Geography readoutThe site's own market map: sessions, engagement, conversions by country.
- Device readoutDevice-class truth from the site's own measurements.
- Events inventoryEvery event the property collects, ranked, orphans flagged.
- Source and medium auditWhether traffic lands in the buckets it should, before channel reports are trusted.
Cross-platform
18 skills- All-platform snapshotOne table: spend and results everywhere, right now.
- Health check sweepWhich connections are alive, degraded or dead, before work starts.
- Account inventory sweepThe full map of ticked accounts everywhere, with settings.
- Pixel audit sweepWhich platforms have working event wiring, in one table.
- Change log sweepEverything humans and systems changed, everywhere, in one timeline.
- Currency and timezone mapThe conversion and alignment rules every merged table needs.
- Executive spend summarySpend by platform and purpose, reconciled, caveated, one page.
- Promo readiness checkThe T-minus checklist across platforms: budgets, creatives, feeds, tracking.
- Seasonal preflightBudget headroom, learning stability and feed health before the season, not during.
- Spend freeze previewThe full stop, staged as Safe Write previews, ready for one human yes.
- Naming consistency auditWhether names encode what reporting needs, everywhere, the same way.
- UTM consistency auditWhether the links ads carry land in GA4 as the channels they claim.
- Catalogue coverage sweepWhich products serve where, the cross-platform commerce map.
- Post-incident verificationProof that everything restarted, nothing forgotten, learning acknowledged.
- Creative sweepThe best and worst performing creative everywhere, one ranked view.
- Morning checkOvernight surprises caught before the client sees them.
- Alert baseline builderWhat 'normal' is, per account and metric, so deviations mean something.
- Data freshness mapThe lag table: what 'today' means on every platform, written down once.
For your agent
Discovery works like the tools do.
Every MCP endpoint announces two skill tools. An assistant finds the right contract by need, then loads it whole, without anyone opening a browser.
search_skills- Finds skills by need, source or tool. Returns names, intents and tiers.
get_skill- Returns one contract in full: procedure, decision rules, evidence, failure modes.
/skills.md- The whole index as Markdown. Every skill also has its own twin at /skills/{slug}.md.