Prompt Guide: 27 Ad Platforms and Support Tools

Use these prompts in the Synter Campaign IDE or through the Synter MCP server. Replace text in [BRACKETS] with your own values, except for operation IDs, configuration hashes, expected before-states, and resource versions: copy those values verbatim from the agent's preflight response into the approval.

Last reviewed: September 3, 2026.

---

Start With a Safe, Complete Prompt

The best prompts name:

1. Account - platform and advertiser/account name or ID
2. Action - report, plan, create, update, pause, enable, or verify
3. Scope - campaign IDs, dates, budget type, geography, and audience
4. Success metric - qualified leads, purchases, CPA, ROAS, reach, or another business outcome
5. Safety boundary - report only, create paused, wait for approval, or execute and read back

"In Google Ads account [ACCOUNT], pull campaign performance from [START DATE] through [END DATE]. Compare spend, qualified conversions, and CPA with our $[TARGET] CPA. Do not change anything. Return no more than five recommended actions with campaign IDs and supporting data."

Operating rules worth including

• Select the intended account in the Campaign IDE's top account selector before asking the agent to act.
• Use exact dates for reporting. Platform presets and time zones do not always cover identical days.
• Ask the agent to list and propose first, then approve every create, upload, sync, update, delete, archive, detach, or cleanup in a second message that repeats the platform connection and immutable account/advertiser ID. The agent must reject a mutation if its current scope differs. Do not combine metric-based selection with a bulk mutation.
• Have the agent generate a unique operation ID during preflight. Copy that ID verbatim into the approval and every retry of that same intended write. Never generate a new ID for a retry or reuse an ID for a different write. Use it as the platform idempotency key where supported; otherwise reconcile by scope, operation ID, and exact entity name/configuration before any retry. If completion remains ambiguous, stop instead of replaying the write.
• For updates, have the preflight return the expected current values and resource version or ETag when available. Repeat them with exact absolute target values in the approval, re-read immediately before writing, and reject stale approvals. Use a conditional update when the platform supports one.
• For every monetary value, have the preflight read the advertising account's ISO 4217 currency and return both the human amount and the platform API units, such as micros. Repeat the currency and units in the approval and reject the write if the live account currency differs. A $ placeholder is valid only after the preflight verifies that the account currency is USD; otherwise replace it with the exact account currency.
• For a multi-entity create, treat every hierarchy level as part of the operation. If only part completes, do not replay or claim rollback. Report every created ID. Resuming or cleaning up requires a new scoped preflight and a second approval; the original create approval authorizes neither action.
• Ask for new campaigns in PAUSED or draft state. Review targeting, creative, destination, budget, and tracking before enabling spend.
• Ask for a before-and-after readback after every update.
• Verify the conversion event on the live destination before using conversion-based bidding.
• Use a staged CSV or audience artifact for customer data. Never paste email addresses, phone numbers, or other PII into chat. To create an artifact, bind the preflight and approval to the source scope and snapshot, proposed artifact name, owning tenant/workspace, retention, expected row count, and schema/field mapping; include them in the configuration hash, then read back the assigned immutable artifact ID, version, and digest. For every later write that consumes an existing artifact, repeat that ID, version, digest, owner, row count, and schema/field mapping in the preflight and approval, include them in the configuration hash, re-read them immediately before the write, and reject any mismatch.
• Refer to uploaded assets by Creative Library name or artifact ID, or use a public HTTPS URL. A path on your computer is not available to the agent.
• Send paid traffic to a content or landing page with UTMs, not directly to a signup or onboarding page.

Reusable write approval sequence

1. Preflight the exact scope and proposed write without changing anything:

"In [PLATFORM] connection [CONNECTION], account/advertiser ID [ACCOUNT ID], validate the current scope and propose [CREATE/UPLOAD/SYNC/UPDATE/DELETE/ARCHIVE/DETACH/CLEANUP] for [EXACT ENTITY OR IDS]. Generate and return a unique operation ID and a hash of the complete proposed configuration. For every monetary value, return the live account's ISO 4217 currency, human amount, and platform API units. Confirm that no matching entity already exists for a create. For an update or destructive action, return the exact expected current values, resource version or ETag when available, dependency impact, and ordered action plan. Do not make changes."

2. Approve that exact write:

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. Exact preflighted configuration (verbatim, not a summary): [FULL PREFLIGHT CONFIGURATION]. In [PLATFORM] connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve [EXACT CREATE/UPLOAD/SYNC/UPDATE/DELETE/ARCHIVE/DETACH/CLEANUP] for [EXACT ENTITY OR IDS] only. Approved account currency and platform monetary units: [ISO CURRENCY / HUMAN AMOUNT / API UNITS, OR NO MONETARY VALUE]. For a create, the expected before-state is that no matching entity exists. For an update or destructive action, the expected before-state and version are [EXPECTED CURRENT STATE AND VERSION], dependency impact is [DEPENDENCIES], and ordered action plan is [ORDERED PLAN]. The exact target state is [TARGET STATE]. Immediately re-read every target and reject this request if its scope, configuration hash, account currency, platform units, dependencies, or before-state differs from the preflight. If NO MONETARY VALUE is declared but any target field is monetary, reject the request. Use the operation ID as an idempotency key where supported. Apply only the approved write, then read every target back and show the platform, connection, account/advertiser ID, resulting IDs, and before-and-after state. If the response is lost, ambiguous, or partial, reconcile by scope, operation ID, and exact entity name/configuration. Do not replay or claim rollback; report every completed write. A resume or cleanup requires a new preflight and separate approval."

Create, upload, and sync prompts that begin with an operation ID and configuration hash are approval payloads for step 2, not standalone prompts. Platform-specific templates abbreviate the reusable payload: before sending one, add every applicable field from steps 1 and 2, including the verified account currency, human monetary amount, platform API units, and staged-artifact metadata. Update prompts that say not to apply a change are step-1 preflights; send a separate approval containing the returned operation ID, configuration hash, scope, expected state/version, and exact target state. Create approvals require the preflight to confirm that no matching entity exists. Every mention of resume or cleanup means a new preflight followed by a separate approval. A write request without the applicable fields is incomplete and must be rejected.

---

Current Platform Coverage

Support depth differs by platform. The prompts below stay within the current Synter surface.

---

Platform Prompt Library

1. Google Ads

Capability: Build. Customer Match, conversion actions, offline conversions, GA4 linking, and advanced GAQL reporting are available.

"In Google Ads account [ACCOUNT], report campaign performance from [START DATE] through [END DATE]. Show spend, impressions, clicks, CTR, conversions, qualified conversions if available, CPA, conversion value, and ROAS. Do not change anything."

"Pull Search terms from [START DATE] through [END DATE]. Return up to 25 terms with at least [N] clicks and zero qualified conversions. Group them by campaign and label each as a negative-keyword candidate. Do not add negatives yet."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Google Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED Google Search campaign named [NAME]. Daily budget: [AMOUNT] [ISO ACCOUNT CURRENCY], represented as [MICROS] micros. Geography: [LOCATION]. Landing page: [CONTENT URL WITH UTMS]. Build tightly themed ad groups for [KEYWORDS], use phrase and exact match, and draft at least three distinct RSA headlines per theme. Verify the primary conversion action before choosing Maximize Conversions. Immediately re-read the scope, account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back the full campaign. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request a new resume-or-cleanup preflight; do not replay or claim rollback."

"In Google Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show the current bid strategy, target, and daily budget for campaign ID [ID]. Propose one exact absolute change based on performance from [START DATE] through [END DATE]. For any monetary value, return the live ISO account currency, human amount, and platform API units. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not apply it. Apply only after my second approval repeats the operation ID, hash, scope, expected before-state/version, exact target state, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

2. Meta Ads (Facebook & Instagram)

Capability: Build. Campaigns, ad sets, ads, custom/lookalike audiences, pixels, and CAPI workflows are available.

"In Meta account [ACCOUNT], pull campaign and ad-level performance from [START DATE] through [END DATE]. Show spend, impressions, CPM, link clicks, CTR, landing-page views, qualified leads or purchases, CPA, and frequency. Flag creative-fatigue candidates, but do not change anything."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Meta connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED website-leads campaign named [NAME]. Daily campaign budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Audience: [DESCRIPTION]. Destination: [CONTENT URL WITH UTMS]. Optimize for [LEAD EVENT] only after verifying that the selected pixel belongs to this account and that the event is firing. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Create and read back the complete campaign, ad set, and ads. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Operation ID [SOURCE OPERATION ID]. Preflight configuration hash [SOURCE CONFIGURATION HASH]. In Meta connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve using staged audience artifact [ARTIFACT NAME/ID] at immutable version [ARTIFACT VERSION] and content digest [SHA-256], owned by tenant/workspace [TENANT/WORKSPACE ID], with row count [ROW COUNT] and schema/field mapping [SCHEMA/MAPPING], to create only a Custom Audience named [SOURCE AUDIENCE NAME]. Immediately re-read the scope and all artifact metadata and confirm no matching source audience exists; reject any scope, configuration-hash, artifact ID, version, digest, owner, row-count, schema, mapping, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Report the source audience ID, processing status, matched size, and lookalike eligibility without displaying PII. Do not create a lookalike, attach the audience, or activate spend. If completion is ambiguous, reconcile the source audience before any retry."

"After a separate read confirms source Custom Audience [SOURCE AUDIENCE ID] is READY and meets Meta's current lookalike eligibility threshold, preflight one 1% lookalike in [COUNTRY]. Return a new agent-generated operation ID and configuration hash, the immutable Meta connection and account ID, source audience ID, readiness evidence, threshold result, country, and proposed lookalike name. Confirm no matching lookalike exists. Do not create it. Create the lookalike only after my second approval repeats every returned value, re-read readiness and eligibility immediately, and rejects any scope, state, threshold, hash, or create-precondition mismatch. Report the resulting lookalike ID and processing state; do not attach it or enable spend. Reconcile an ambiguous result before retrying."

"Operation ID [LOOKALIKE OPERATION ID]. Preflight configuration hash [LOOKALIKE CONFIGURATION HASH]. In Meta connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating only the preflighted 1% lookalike named [LOOKALIKE NAME] in [COUNTRY] from source Custom Audience [SOURCE AUDIENCE ID], which was READY with matched size [SIZE] against eligibility threshold [THRESHOLD] at [CHECK TIME]. Immediately re-read the source and reject any scope, hash, readiness, size, threshold, country, name, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Report the resulting lookalike ID and processing state; do not attach it or enable spend. Reconcile an ambiguous result before retrying."

"In Meta connection [CONNECTION], account/advertiser ID [ACCOUNT ID], list up to five campaigns above [CPA] [ISO ACCOUNT CURRENCY] CPA with at least [MINIMUM SPEND] [ISO ACCOUNT CURRENCY] spend from [START DATE] through [END DATE]. Show current budgets and child delivery status. For each proposed absolute change, return the live ISO account currency, human amount, platform API units, an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not pause or resize anything. Apply only after my second approval repeats each operation ID, hash, scope, exact campaign ID, expected before-state/version, exact target state, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use conditional updates and idempotency keys where supported, and show the before-and-after hierarchy. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

3. LinkedIn Ads

Capability: Build. Campaign Groups, Campaigns, Creatives, Lead Gen Forms, matched audiences, and conversion rules are available.

"In LinkedIn account [ACCOUNT], pull campaign-level results from [START DATE] through [END DATE]. Show spend, impressions, clicks, CTR, leads, CPL, and lead-form submissions. Do not infer demographic performance that the report does not expose."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In LinkedIn connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED lead-generation build named [NAME]: one Campaign Group, [CAMPAIGN COUNT] Campaigns grouped by [THEME], and [CREATIVES PER CAMPAIGN] Creatives per Campaign. Target [JOB FUNCTIONS/TITLES], [SENIORITIES], [INDUSTRIES], and [GEOGRAPHY], and exclude [WRONG-FIT JOB FUNCTIONS]. Attach Lead Gen Form [FORM] and conversion rule [RULE]. Immediately re-read the scope and confirm no matching hierarchy exists; reject any scope, configuration-hash, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back the complete hierarchy. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In LinkedIn connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve uploading staged CSV artifact [ARTIFACT NAME/ID] at immutable version [ARTIFACT VERSION] and content digest [SHA-256], owned by tenant/workspace [TENANT/WORKSPACE ID], with row count [ROW COUNT] and schema/field mapping [SCHEMA/MAPPING], as a matched audience named [NAME]. Immediately re-read the scope and all artifact metadata and confirm no matching audience exists; reject any scope, configuration-hash, artifact ID, version, digest, owner, row-count, schema, mapping, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Do not paste or display member PII. Report the scope, audience ID, processing status, and matched size when available. Do not activate a campaign until the audience is READY and large enough to serve. If completion is ambiguous, reconcile by scope, operation ID, name, and configuration; do not replay unless absence is proven."

"List LinkedIn campaigns with more than $[MINIMUM SPEND] spend and fewer than [N] leads from [START DATE] through [END DATE]. Include Campaign Group status and exact IDs. Recommend actions only."

4. X Ads (Twitter)

Capability: Build. Campaigns, line items, promoted posts, Website Cards, tailored audiences, web tags, and reporting are available.

"In X Ads account [ACCOUNT], report campaign and line-item performance from [START DATE] through [END DATE] in the account time zone. Show spend, impressions, engagements, clicks, conversions, CPE, and CPC. Do not use a relative date preset."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In X Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED website-traffic build named [NAME]: campaign shell, line item at [BUDGET] [ISO ACCOUNT CURRENCY] per day, represented as [PLATFORM API UNITS], and Website Card for [CONTENT URL WITH UTMS], targeting [INTERESTS/KEYWORDS/FOLLOWER LOOKALIKES]. Keep the promoted-post copy free of URLs and hashtags, with no more than one emoji. Immediately re-read the scope, live account currency, and monetary units and confirm no matching build exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back the campaign and line-item statuses. If completion is ambiguous or partial, reconcile the build, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In X Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve using staged audience artifact [ARTIFACT NAME/ID] at immutable version [ARTIFACT VERSION] and content digest [SHA-256], owned by tenant/workspace [TENANT/WORKSPACE ID], with row count [ROW COUNT] and schema/field mapping [SCHEMA/MAPPING], to create a tailored audience named [NAME]. Immediately re-read the scope and all artifact metadata and confirm no matching audience exists; reject any scope, configuration-hash, artifact ID, version, digest, owner, row-count, schema, mapping, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Report the scope, audience ID, size, and eligibility. Do not attach it or enable spend until it meets X's delivery threshold. If completion is ambiguous, reconcile by scope, operation ID, name, and configuration; do not replay unless absence is proven."

"In X Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show the current budget and status for line item ID [ID] and propose an exact daily budget of [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, line-item ID, expected before-state/version, exact target budget, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

5. Reddit Ads

Capability: Build. Complete campaign hierarchies, community/interest targeting, custom audiences, pixel inspection, CAPI, and reporting are available.

"In Reddit Ads account [ACCOUNT], pull campaign and ad-group performance from [START DATE] through [END DATE]. Show spend, impressions, clicks, CTR, conversions, CPA, and current delivery status."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Reddit Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one complete PAUSED conversion campaign named [NAME]. Target communities [LIST] and [LOCATION]. Use approved image creative [CREATIVE ASSET] and [CONTENT URL WITH UTMS]. Optimize for the verified [EVENT] pixel event with an ad-group budget of [BUDGET] [ISO ACCOUNT CURRENCY] per day, represented as [PLATFORM API UNITS]. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Create campaign, ad group, post, and ad in deterministic order, all paused; stop on the first failure and read back every created ID. If completion is ambiguous or partial, reconcile by scope, operation ID, name, and configuration, then report whether an explicit cleanup approval or a resume approval is required; do not replay or claim rollback."

"In Reddit Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show which pixel events have fired, then prepare one Lead test event with unique event ID [EVENT ID] using Reddit's test-event mode. Return an agent-generated operation ID and configuration hash. Do not send it. Send only after my second approval repeats the operation ID, hash, connection, account, pixel, event ID, and exact test payload. Re-read and reject mismatches. Never send a test conversion to a production event stream when test mode is unavailable. If completion is ambiguous, reconcile by operation ID and event ID before retrying."

"List up to five Reddit ad groups above $[CPA] CPA and $[MINIMUM SPEND] spend from [START DATE] through [END DATE]. Include campaign, ad-group, and ad IDs. Recommend actions only."

6. Microsoft / Bing Ads

Capability: Build. Search campaigns, ad groups, audiences, negatives, schedules, UET conversions, reporting, and structure validation are available.

"In Microsoft Ads account [ACCOUNT], pull campaign performance from [START DATE] through [END DATE]. Show spend, impressions, clicks, CTR, conversions, CPA, and current daily budget."

"Pull Microsoft Ads search terms from [START DATE] through [END DATE]. Return up to 25 non-converting terms with at least $[MINIMUM SPEND] spend, grouped by campaign. Do not add negatives yet."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Microsoft Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED Search campaign named [NAME]. Daily budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Geography: [LOCATION]. Keywords: [LIST]. Landing page: [CONTENT URL WITH UTMS]. Verify the UET goal before using conversion bidding. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back the campaign and ad groups. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"In Microsoft Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show the current status and budget for campaign ID [ID] and propose changing [FIELD] to exact value [VALUE]. For any monetary value, return the live ISO account currency, human amount, and platform API units. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, exact target value, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after hierarchy. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

7. TikTok Ads

Capability: Build. Campaigns, ad groups, ads, Spark Ads, Shop/catalog campaigns, audiences, pixels, Events API tests, and reporting are available.

"In TikTok Ads account [ACCOUNT], pull campaign, ad-group, and ad performance from [START DATE] through [END DATE]. Show spend, impressions, six-second views, completed views, CTR, conversions, CPA, and creative-level results."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In TikTok Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED web-conversion campaign named [NAME]. Daily budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Audience: [DESCRIPTION]. Video asset: [CREATIVE LIBRARY NAME/ID OR HTTPS URL]. Destination: [CONTENT URL WITH UTMS]. Verify that pixel event [EVENT] has fired before selecting conversion optimization. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Build and read back the campaign, ad group, and ad. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In TikTok Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED Spark Ad for authorized post [POST ID] targeting [AUDIENCE] in [LOCATION], with [BUDGET] [ISO ACCOUNT CURRENCY] per day, represented as [PLATFORM API UNITS], and destination [CONTENT URL WITH UTMS]. Validate identity authorization. Immediately re-read the scope, live account currency, and monetary units and confirm no matching ad exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Show the full hierarchy before activation. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In TikTok Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve using staged audience artifact [ARTIFACT NAME/ID] at immutable version [ARTIFACT VERSION] and content digest [SHA-256], owned by tenant/workspace [TENANT/WORKSPACE ID], with row count [ROW COUNT] and schema/field mapping [SCHEMA/MAPPING], to create a Custom Audience named [NAME]. Immediately re-read the scope and all artifact metadata and confirm no matching audience exists; reject any scope, configuration-hash, artifact ID, version, digest, owner, row-count, schema, mapping, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Report the scope, audience ID, processing status, and size. Do not attach or enable it until it is valid and eligible to serve. If completion is ambiguous, reconcile by scope, operation ID, name, and configuration; do not replay unless absence is proven."

8. Amazon DSP

Capability: Build. Advertisers, campaigns/orders, ad groups/line items, audiences, domain targeting, and asynchronous reporting are available.

"For Amazon DSP advertiser [ADVERTISER], pull performance from [START DATE] through [END DATE], no more than 90 days. Show spend, impressions, clicks, CTR, video completions, conversions, and ROAS by campaign and line item."

"List every active Amazon DSP campaign for advertiser [ADVERTISER]. Show campaign ID, budget, flight dates, status, and child line-item count. Paginate the full result."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Amazon DSP connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one INACTIVE campaign named [NAME]. Lifetime budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Flight: [START DATE] through [END DATE]. Objective: [OBJECTIVE]. Add INACTIVE line items for [SUPPLY/AUDIENCE PLAN]. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back the hierarchy and verify every entity is INACTIVE. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"In Amazon DSP connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show the current flight, budget, and status for campaign [ID] and propose exact absolute [BUDGET/END DATE] values. For a budget change, return the live ISO account currency, human amount, and platform API units. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, exact target values, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after campaign and affected line items. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

9. Amazon Sponsored Ads

Capability: Build. Sponsored Products, Sponsored Brands, Sponsored Display, portfolios, keyword/product targets, recommendations, search terms, and reporting are available.

"For Amazon Ads profile [PROFILE], report Sponsored Products, Sponsored Brands, and Sponsored Display separately from [START DATE] through [END DATE], no more than 90 days. Show spend, sales, orders, ACOS, ROAS, and conversion rate by campaign."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Amazon Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED manual-keyword Sponsored Products campaign for ASIN [ASIN]. Name: [NAME]. Daily budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [BUDGET API UNITS]. Exact keywords: [LIST]. Default bid: [BID] [ISO ACCOUNT CURRENCY], represented as [BID API UNITS]. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Create and read back the campaign, ad group, product ad, and keyword targets. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Amazon Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one separate PAUSED product-targeting Sponsored Products campaign for ASIN [ASIN]. Target these ASINs or categories: [LIST]. Daily budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Do not mix keyword and product targets in the same campaign. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back the complete structure. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Show Amazon's bid recommendations for campaign [ID] and compare them with current bids. Do not apply bid changes. Return a review table with estimated impact and risk."

10. Pinterest Ads

Capability: Build. Campaigns, ad groups, promoted Pins, budget updates, status changes, and reporting are available.

"In Pinterest Ads account [ACCOUNT], pull campaign and ad performance from [START DATE] through [END DATE]. Show spend, impressions, saves, outbound clicks, CTR, conversions, and CPA."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Pinterest Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED campaign named [NAME] with objective [OBJECTIVE], daily budget [BUDGET] [ISO ACCOUNT CURRENCY] represented as [PLATFORM API UNITS], targeting [AUDIENCE/KEYWORDS/LOCATION], existing Pin [PIN ID], and destination [CONTENT URL WITH UTMS]. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back the campaign, ad group, and ad status. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"In Pinterest Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show the current budget and delivery status for campaign [ID] and propose an exact daily budget of [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, exact target budget, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

11. Snapchat Ads

Capability: Build. Campaigns, ad squads, media, creatives, ads, pixels, test events, and reporting are available.

"In Snapchat Ads account [ACCOUNT], report campaign, ad-squad, and ad performance from [START DATE] through [END DATE]. Show spend, impressions, swipe-ups, CTR, conversions, CPA, and video completion metrics."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Snapchat Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED campaign named [NAME]. Objective: [OBJECTIVE]. Daily budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Audience: [DESCRIPTION]. Use approved vertical asset [CREATIVE NAME/ID OR HTTPS URL] and [CONTENT URL WITH UTMS]. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Build and read back the campaign, ad squad, creative, and ad. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"In Snapchat Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], list the pixels, then prepare one [EVENT] test event with unique event ID [EVENT ID] for pixel [PIXEL ID] using Snapchat's test-event mode. Return an agent-generated operation ID and configuration hash. Do not send it. Send only after my second approval repeats the operation ID, hash, connection, account, pixel, event ID, and exact test payload. Re-read and reject mismatches. Never send a test conversion to a production event stream when test mode is unavailable. If completion is ambiguous, reconcile by operation ID and event ID before retrying."

"In Snapchat Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show the current budget and child status for campaign [ID] and propose an exact budget of [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS], with the intended child-status cascade. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, exact target budget, currency, human amount, API units, and status action. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after hierarchy. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

12. Spotify Ads

Capability: Build. Campaigns, ad sets, audio ads, geo/topic targeting, budget updates, status changes, and reporting are available.

"In Spotify Ads account [ACCOUNT], report performance from [START DATE] through [END DATE]. Show spend, impressions, reach, clicks, CTR, audio completions, and completion rate by campaign and ad set."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Spotify Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED audio campaign named [NAME]. Objective: [REACH/CLICKS/WEB TRAFFIC]. Ad-set budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Target [AGE/GENDER/LOCATION/GENRES/INTERESTS]. Use approved audio [ASSET NAME/ID OR HTTPS URL], companion image [ASSET NAME/ID OR HTTPS URL], and [CONTENT URL WITH UTMS]. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back the campaign, ad set, and ad. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Spotify Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED podcast campaign targeting podcast topics [TOPICS], not individual shows. Preflight-confirmed budget type: [LIFETIME OR DAILY — COPY THE ONE RETURNED]. Exact budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Use approved audio and companion image assets. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, budget-type, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back the draft structure and exact budget type. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"In Spotify Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show the current ad-set budget for campaign [ID] and propose exactly [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, exact target budget, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

13. OpenAI Ads / LLM Placements

Capability: Build. Campaigns, ad groups, ads, creative upload, conversion-event settings, activation, and reporting are available.

"In OpenAI Ads account [ACCOUNT], report campaign, ad-group, and ad performance from [START DATE] through [END DATE]. Show spend, impressions, clicks, conversions, CPA, and the billing event used by each campaign."

"List the conversion-event settings and conversion data sources available to OpenAI Ads account [ACCOUNT]. Verify CAPI readiness before proposing conversion-optimized bidding."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In OpenAI Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED campaign named [NAME] with a lifetime budget cap of [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS], for [COUNTRIES]. Billing event: [IMPRESSION OR CLICK]. Bidding objective: [IMPRESSIONS/CLICKS/CONVERSIONS]. Use creative [ASSET NAME/ID]. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Create compatible ad groups and ads, then read back billing configuration, bids, budget, and statuses. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"In OpenAI Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show the current lifetime budget cap and status for campaign [ID] and propose exact [FIELD] = [VALUE] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it or convert the lifetime cap into a daily budget. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, field, exact target value, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

14. The Trade Desk (TTD)

Capability: Build. Advertisers, campaigns, ad groups, first-party audiences, budget updates, status changes, and reporting are available.

"For The Trade Desk advertiser [ADVERTISER], report performance from [START DATE] through [END DATE]. Show spend, impressions, clicks, CTR, conversions, CPA, and pacing by campaign."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In The Trade Desk connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one non-serving campaign named [NAME]. Budget: [BUDGET] [ISO ACCOUNT CURRENCY] [TOTAL/DAILY], represented as [PLATFORM API UNITS]. Flight: [START DATE] through [END DATE]. Objective: [OBJECTIVE]. Add ad groups for [AUDIENCE/CHANNEL PLAN] with every ad group's IsEnabled=false. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back every created ID and IsEnabled value. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In The Trade Desk connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve using staged artifact [ARTIFACT NAME/ID] at immutable version [ARTIFACT VERSION] and content digest [SHA-256], owned by tenant/workspace [TENANT/WORKSPACE ID], with row count [ROW COUNT] and schema/field mapping [SCHEMA/MAPPING], to create a first-party audience named [NAME]. Immediately re-read the scope and all artifact metadata and confirm no matching audience exists; reject any scope, configuration-hash, artifact ID, version, digest, owner, row-count, schema, mapping, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Report the audience ID, upload status, and match status without displaying member data. Do not activate it. If completion is ambiguous, reconcile by scope, operation ID, name, and configuration; do not replay unless absence is proven."

"In The Trade Desk connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show current budget type, budget value, pacing, and status for campaign [ID] and propose exact [TOTAL BUDGET/DAILY BUDGET] = [VALUE] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it or convert one budget type into the other. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, budget field, exact target value, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

15. FreeWheel (CTV & Video)

Capability: Report only. Synter does not currently create or change FreeWheel campaigns.

"For my selected FreeWheel advertiser, pull performance from [START DATE] through [END DATE]. Show spend, impressions, clicks, CTR, video starts, video completions, VCR, and conversions by campaign. Confirm the advertiser scope used."

"Compare FreeWheel campaign performance with DV360 and The Trade Desk from [START DATE] through [END DATE]. Normalize currency and date boundaries, and explain any metrics that are not directly comparable. Do not change anything."

"Check my FreeWheel connection and advertiser scope. List the reporting operations currently available. Do not invent campaign, bid, budget, or creative writes."

16. Display & Video 360 (DV360)

Capability: Build. Campaigns, insertion orders, line items, flight/budget updates, status changes, structure validation, and reporting are available.

"For DV360 advertiser [ADVERTISER], list all campaigns, insertion orders, and line items. Show IDs, status, budget, pacing, flight dates, and hierarchy. Paginate the full result."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In DV360 connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one DRAFT campaign named [NAME], insertion order [IO NAME] with a total budget of [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS], from [START DATE] through [END DATE], and draft line items for [DISPLAY/VIDEO/CTV PLAN]. Immediately re-read the scope, live account currency, and monetary units and confirm no matching hierarchy exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back every entity and verify none is serving. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"In DV360 connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show the current flight, budget segments, pacing, and status for insertion order [ID] and propose exact absolute [BUDGET/START DATE/END DATE/PACING] values. For any monetary value, return the live ISO account currency, human amount, and platform API units. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, insertion-order ID, expected before-state/version, exact target values, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after insertion order and line items. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

"Validate DV360 campaign [ID]. Flag missing creatives, invalid flight nesting, inactive parents, and line items that cannot serve. Recommend fixes only."

17. Campaign Manager 360 (CM360)

Capability: Report only. Use DV360 for campaign, insertion-order, or line-item changes.

"In CM360 account [ACCOUNT], pull performance from [START DATE] through [END DATE]. Show impressions, clicks, conversions, media cost, and placement-level results where available. Do not change anything."

"Compare CM360 ad-serving totals with DV360 delivery from [START DATE] through [END DATE]. Reconcile date, currency, attribution, and placement differences. Flag gaps without assuming either source is wrong."

"Check the selected CM360 profile and list the current reporting capability. Do not invent placement, creative, ad, or Floodlight mutations."

18. Walmart Connect

Capability: Manage. Reporting, campaign pause/enable, and daily-budget updates are available; complete campaign creation is not.

"Pull Walmart Connect campaign performance from [START DATE] through [END DATE]. Show spend, impressions, clicks, attributed conversions, attributed sales, ROAS, and current daily budget."

"List up to five Walmart campaigns with ROAS below [TARGET] and spend above $[MINIMUM] from [START DATE] through [END DATE]. Show exact IDs and current budgets. Recommend actions only."

"In Walmart Connect connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show current status and daily budget for campaign [ID] and propose exact action [PAUSE/ENABLE/SET DAILY BUDGET TO AMOUNT]. For a budget change, return the live ISO account currency, human amount, and platform API units. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, exact target state, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

19. Instacart Ads

Capability: Manage (contract-dependent). Reporting is available; pause/enable and daily-budget writes require the connected partner contract. Verify the live connection's write operations before preparing an approval.

"Pull Instacart Ads performance from [START DATE] through [END DATE]. Show impressions, clicks, spend, attributed orders, attributed sales, ROAS, and current budget by campaign."

"List up to five Instacart campaigns with spend above $[MINIMUM] and attributed orders below [N] from [START DATE] through [END DATE]. Recommend actions only."

"In Instacart Ads connection [CONNECTION], account/advertiser ID [ACCOUNT ID], verify that this connection exposes the required partner write contract for campaign [ID]. Show current status and daily budget and propose exact action [PAUSE/ENABLE/SET DAILY BUDGET TO AMOUNT]. For a budget change, return the live ISO account currency, human amount, and platform API units. If supported, return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, exact target state, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied. If unsupported, report the exact boundary without guessing."

20. Target Roundel

Capability: Manage (contract-dependent). Reporting is available; pause/enable and daily-budget writes require the connected partner contract. Verify the live connection's write operations before preparing an approval.

"Pull Target Roundel performance from [START DATE] through [END DATE]. Show spend, impressions, clicks, attributed conversions, attributed sales, ROAS, and current budget by campaign."

"Compare Roundel onsite and offsite results from [START DATE] through [END DATE] where those dimensions are available. Identify up to five budget-review candidates and do not change anything."

"In Target Roundel connection [CONNECTION], account/advertiser ID [ACCOUNT ID], verify that this connection exposes the required partner write contract for campaign [ID]. Show current status and daily budget and propose exact action [PAUSE/ENABLE/SET DAILY BUDGET TO AMOUNT]. For a budget change, return the live ISO account currency, human amount, and platform API units. If supported, return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, exact target state, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied. If unsupported, report the exact boundary without guessing."

21. Criteo Commerce Media

Capability: Manage. Reporting, campaign pause/enable, and budget updates are available; complete campaign creation is not.

"Pull Criteo Commerce Media performance from [START DATE] through [END DATE]. Show displays, clicks, spend, conversions, revenue, conversion rate, and ROAS by campaign."

"List up to five Criteo campaigns below [TARGET ROAS] with spend above $[MINIMUM] from [START DATE] through [END DATE]. Show exact IDs and current budgets. Recommend actions only."

"In Criteo connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show current status and budget for campaign [ID] and propose exact action [PAUSE/ENABLE/SET BUDGET TO AMOUNT]. For a budget change, return the live ISO account currency, human amount, and platform API units. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, exact target state, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

22. StackAdapt

Capability: Build in the Campaign IDE; Manage through the hosted MCP. Reporting, campaign/ad changes, audiences, inventory discovery, and conversion pixels are available in both where exposed.

"For StackAdapt advertiser [ADVERTISER], pull TOTAL campaign performance from [START DATE] through [END DATE]. Show spend, impressions, clicks, conversions, eCPM, eCPC, CTR, and pacing. Also list the ads attached to campaign [ID]."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In StackAdapt connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve building one non-serving [NATIVE/DISPLAY/VIDEO/CTV/AUDIO/DOOH] campaign named [NAME] in the Campaign IDE. Flight: [START DATE] through [END DATE]. Budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Use preflighted inventory, audience, geo, conversion pixel [PIXEL ID], creative [ASSET IDS], and ads. Immediately re-read the scope, live account currency, and monetary units and confirm no matching campaign exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Verify every expected ad by readback and do not enable spend. If completion is ambiguous or partial, reconcile the hierarchy, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In StackAdapt connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve using staged artifact [ARTIFACT NAME/ID] at immutable version [ARTIFACT VERSION] and content digest [SHA-256], owned by tenant/workspace [TENANT/WORKSPACE ID], with row count [ROW COUNT] and schema/field mapping [SCHEMA/MAPPING], to create an [EMAIL/DOMAIN ABM] audience named [NAME]. Immediately re-read the scope and all artifact metadata and confirm no matching audience exists; reject any scope, configuration-hash, artifact ID, version, digest, owner, row-count, schema, mapping, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Report the audience ID and matching status without displaying member data. Do not attach it yet. If completion is ambiguous, reconcile by scope, operation ID, name, and configuration; do not replay unless absence is proven."

"In StackAdapt connection [CONNECTION], account/advertiser ID [ACCOUNT ID], show the current flight budget, attached conversion pixels, and status for campaign [ID] and propose exact [FIELD] = [VALUE]. For any monetary value, return the live ISO account currency, human amount, and platform API units. Return an agent-generated operation ID and configuration hash, expected current values, and resource version or ETag. Do not change it or treat a retargeting pixel as a conversion tracker. Apply only after my second approval repeats the operation ID, hash, scope, campaign ID, expected before-state/version, field, exact target value, currency, human amount, and API units. Re-read immediately and reject any scope, hash, currency, units, or before-state mismatch; use a conditional update and idempotency key where supported, and show the before-and-after readback. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

23. Search Ads 360 (SA360)

Capability: Report, limited to campaign inventory and campaign verification. Synter does not pull SA360 metrics or mutate campaigns.

"List every campaign visible to selected SA360 customer [CUSTOMER]. Show campaign name, engine campaign ID, status, and the identifiers needed to reconcile it with Google Ads. Paginate the full inventory."

"Verify whether Google Ads campaign [EXACT NAME OR ENGINE ID] is visible in SA360 after the change made on [DATE]. Return visible, pending, or not found, and explain the next check."

"Check my selected SA360 reporting customer and list the exact read operations available. Do not claim performance reporting, budget changes, pauses, or campaign creation."

24. Samsung Ads

Capability: Alias through DV360. Every Samsung action must be scoped to Samsung inventory available through the connected DV360 advertiser.

"For DV360 advertiser [ADVERTISER], identify campaigns and line items buying Samsung Ads inventory. Show status, flight, budget, pacing, impressions, completed views, and frequency where available."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In DV360 connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one DRAFT Samsung CTV line item under campaign [CAMPAIGN ID] and insertion order [IO ID]. Flight: [START DATE] through [END DATE]. Budget: [BUDGET] [ISO ACCOUNT CURRENCY], represented as [PLATFORM API UNITS]. Use approved CTV video [ASSET ID] and Samsung inventory [DEAL/PACKAGE ID]. Immediately re-read the scope, live account currency, and monetary units and confirm no matching line item exists; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Read back the line item and verify it is not serving. If completion is ambiguous or partial, reconcile the line item and any dependent entity, report every created ID, and request explicit resume-or-cleanup approval; do not replay or claim rollback."

"Validate the Samsung inventory line item [ID] through DV360. Check parent status, deal or inventory assignment, creative approval, flight nesting, frequency cap, and budget. Recommend fixes only."

25. Taboola

Capability: Planned for live campaign management. Synter can forward supported S2S conversions when the Taboola click ID (tblci) is captured.

"Check the current Taboola capability in this workspace. If it is still planned, do not invent account, campaign, audience, or reporting actions. Give me the exact available support and the prerequisites for a future connection."

"Audit [LANDING PAGE URL] for Taboola attribution readiness. Confirm whether tblci is captured and preserved through conversion, whether stable event IDs are available for deduplication, and whether the intended conversion names can be forwarded server-side. Do not send a test production conversion."

"Prepare a Taboola launch package without publishing: audience hypothesis, native headlines, image requirements, destination URL with UTMs, conversion-event map, budget test plan, and a checklist of live API actions that remain unavailable."

26. Outbrain

Capability: Planned. Use Synter for planning and creative preparation, not live account operations.

"Check the current Outbrain capability in this workspace. If it is still planned, do not claim that an advertiser account is connected or that campaigns can be read or changed."

"Prepare an Outbrain native-campaign package without publishing: objective, audience, publisher-safety requirements, headline variants, image specifications, destination URL with UTMs, conversion plan, test budget, and success criteria."

"Adapt approved creative [ASSET ID] into an Outbrain-ready brief. Preserve the offer, propose native headlines and image crops, and flag every field that still requires account-specific API validation."

27. Apple Search Ads

Capability: Planned. Use Synter for App Store search planning and creative preparation, not live account operations.

"Check the current Apple Search Ads capability in this workspace. If it is still planned, do not claim that campaigns, ad groups, keywords, or reports can be accessed."

"Prepare an Apple Search Ads launch plan without publishing for app [APP NAME/ID]. Include Search Results and Search Tab roles, brand/category/competitor keyword groups, exact and discovery structure, negative-keyword strategy, custom product page alignment, test budget, and target cost per acquisition."

"Review the App Store product page assets for [APP]. Map screenshots, app previews, and custom product pages to these keyword themes: [THEMES]. Return an ad-variation brief and flag anything that requires a live Apple Search Ads connection."

---

Support Tool Prompt Library

Support integrations do not count toward the 27 ad platforms. They connect campaign execution to account selection, measurement, customer data, creative production, commerce, and reporting.

Analytics and Funnel Analysis

"Compare GA4 sessions, engaged sessions, key events, and revenue by source / medium and campaign for [DATE RANGE]. Flag paid campaigns where traffic is rising but engagement or conversion rate is falling."

"Use Google Search Console to compare paid-search themes with organic queries for [DATE RANGE]. Find high-impression organic queries that could become paid keyword tests, and paid themes where stronger organic landing-page content may improve relevance."

"Use PostHog to analyze the [FUNNEL NAME] funnel for paid traffic over [DATE RANGE]. Break down drop-off by UTM source, medium, and campaign, then recommend the three highest-impact product or landing-page experiments. Treat CRM-qualified pipeline and revenue—not product events alone—as the source of truth for B2B business outcomes."

Account and Connection State

"List every connected platform and advertiser account in this workspace. Show which accounts are selected in this Campaign IDE thread. Do not expose credentials or tokens."

"I cannot see [ACCOUNT/CAMPAIGN]. First check the top account selection and current account metadata, then test a read-only list operation. Diagnose routing and permissions before suggesting reauthentication."

"Check connection health for [PLATFORM/ACCOUNT]. Distinguish an expired token from a missing parameter, wrong account, insufficient platform role, or transient API error. Do not tell me to reconnect unless token refresh is verified to be impossible or new scopes are required."

Campaign Planning, Preflight, and Readback

"Create a cross-platform campaign plan for [GOAL] across [PLATFORMS]. Total monthly budget: $[BUDGET]. Audience: [DESCRIPTION]. Destination: [CONTENT URL]. Show platform-specific objectives, budget allocation, tracking requirements, creative specs, and risks. Do not create campaigns."

"Run launch preflight for plan [PLAN/NAME]. Verify selected accounts, budget currency and type, destination and UTMs, conversion events, audience eligibility, creative approvals, and every parent/child entity required to serve. Return blockers only; do not enable anything."

"Read back campaign [ID] on [PLATFORM]. Show the complete hierarchy, configured and effective statuses, budget, flight, targeting, creative attachment, conversion tracking, and any policy or delivery blocker."

Cross-Platform Reporting and Budget Pacing

"Pull Google, Meta, LinkedIn, and Microsoft performance from [START DATE] through [END DATE]. Normalize currency and date boundaries. Show spend, qualified conversions, CPA, and CRM opportunities by platform. Explain metrics that are not comparable."

"Compare planned budget with actual spend from [START DATE] through [END DATE]. Show over- and under-pacing by platform and campaign. Recommend moves, but do not change budgets."

"Create a weekly operator report with an executive summary, channel table, material changes, tracking warnings, and no more than five prioritized actions. Link every recommendation to the source metric and date window."

Attribution and Measurement

"Compare ad-platform conversions, GA4 conversions, and CRM-qualified outcomes from [START DATE] through [END DATE]. Quantify the gaps, explain likely attribution differences, and recommend measurement fixes. Do not pause campaigns based only on a source mismatch."

"Show first-touch, last-touch, and linear attribution for [DATE RANGE]. Break results down by platform and campaign, and connect spend to opportunities and revenue where CRM data exists."

"Design an incrementality test for [CAMPAIGN/CHANNEL]. Define treatment, holdout, duration, minimum detectable effect, contamination risks, and the decision rule. Do not launch the test."

"Using at least [N] months of complete spend and revenue data, prepare an MMM scenario for a $[BUDGET] monthly budget. Show assumptions, uncertainty, response curves, and the allocation difference from last-click reporting."

Tracking, Pixels, GTM, and UTMs

"Verify the full conversion path on [LIVE URL]: browser event, Synter receipt, and delivery to [PLATFORMS]. Confirm the event fires only after the intended action. Return evidence for each hop and do not declare success from page-source inspection alone."

"Audit the live GTM container [CONTAINER ID] for [SITE]. List relevant tags, triggers, conversion linker, conflicts, and unpublished changes. Propose the smallest exact fix and return an agent-generated operation ID and configuration hash, expected workspace version, and target state. Do not change or publish the container. Apply or publish only after my second approval repeats the operation ID, hash, account, container ID, workspace, expected version, and exact target state. Re-read immediately, reject mismatches, and reconcile an ambiguous result before retrying."

"In [PLATFORM] connection [CONNECTION], account/advertiser ID [ACCOUNT ID], audit UTM coverage for active campaigns. Show missing or inconsistent utm_source, utm_medium, utm_campaign, utm_content, and utm_term values and propose a canonical suffix. Return an agent-generated operation ID and configuration hash, exact campaign IDs, current suffixes, and resource versions or ETags. Do not apply it. Apply only after my second approval repeats the operation ID, hash, scope, campaign IDs, expected before-state/version, and exact suffix. Re-read immediately, reject mismatches, read back every final URL, and reconcile an ambiguous result before retrying."

"For each destination, preflight only this scope: [PLATFORM / CONNECTION / ACCOUNT OR ADVERTISER ID / PIXEL OR CONVERSION ID]. Return one agent-generated operation ID and configuration hash per destination, the expected current binding, and the exact target binding. Do not configure or send anything. Configure only after my second approval repeats every operation ID, hash, scope, expected binding, and exact target binding. Re-read and reject mismatches, then read back every binding. Prepare one unique test event per destination but send it only after a separate approval repeats the same scope and event IDs; never send test conversions to production event streams or create duplicate conversion events. Reconcile ambiguous results before any retry."

Audiences and CRM

"In [ATTIO/HUBSPOT/SALESFORCE/SHOPIFY] connection [CONNECTION], tenant/workspace/account [SOURCE SCOPE], preflight exporting contacts matching [EXACT FILTER] into encrypted staged artifact [ARTIFACT NAME]. Return the source snapshot/version or cutoff, selected field schema and mapping, expected row count, owning destination tenant/workspace, retention policy, agent-generated operation ID, and configuration hash. Do not export or display PII."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In [ATTIO/HUBSPOT/SALESFORCE/SHOPIFY] connection [CONNECTION], tenant/workspace/account [SOURCE SCOPE], I approve exporting the preflighted source snapshot [VERSION/CUTOFF] with exact filter [FILTER], expected row count [COUNT], schema/field mapping [SCHEMA/MAPPING], destination owner [TENANT/WORKSPACE], artifact name [NAME], and retention [POLICY]. Re-read the source scope and snapshot metadata; reject any scope, hash, version, cutoff, count, schema, mapping, owner, retention, or create-precondition mismatch. Use the operation ID as an idempotency key where supported. Create only the encrypted staged artifact, then report its immutable ID, version, digest, row count, and field coverage without displaying PII. Reconcile an ambiguous result before retrying and do not create a duplicate artifact."

"Operation IDs [ONE AGENT-GENERATED OPERATION ID PER DESTINATION]. Preflight configuration hashes [ONE HASH PER DESTINATION]. For each destination use only this approved scope: [PLATFORM / CONNECTION / ACCOUNT OR ADVERTISER ID]. I approve syncing staged audience [ARTIFACT NAME/ID] at immutable version [ARTIFACT VERSION] and content digest [SHA-256], owned by tenant/workspace [TENANT/WORKSPACE ID], with row count [ROW COUNT] and schema/field mapping [SCHEMA/MAPPING], and creating one separate native audience on each listed platform. Immediately re-read each scope and all artifact metadata and confirm no matching audience exists; reject any scope, configuration-hash, artifact ID, version, digest, owner, row-count, schema, mapping, or create-precondition mismatch. Use each operation ID as the idempotency key where supported. Report each platform's scope, audience ID, processing state, match size, and eligibility. Do not attach or activate anything. If completion is ambiguous, reconcile the affected destination by scope, operation ID, name, and configuration; do not replay unless absence is proven."

"Build an ABM audience for [ICP] using company domains and buying-intent signals. Separate Tier 1 named accounts, Tier 2 vertical clusters, and Tier 3 broader ICP. Show the evidence for each segment before syncing it."

"Check all processing audiences on [PLATFORMS]. Identify ready, too-small, stalled, and failed audiences. Recommend the next action per platform without retrying terminal failures."

Creative Generation and Testing

"Extract the brand guidelines from [WEBSITE] and summarize approved colors, fonts, logos, voice, claims, and prohibited treatments. I will review them before creative generation."

"Audit my existing [PLATFORM] creatives from [START DATE] through [END DATE]. Compare spend, CTR, conversion rate, CPA, and frequency. Identify fatigue and message patterns without assuming correlation proves causation."

"Generate [N] [IMAGE/VIDEO/AUDIO] ad concepts for [PLATFORM] using approved brand [BRAND] and offer [OFFER]. Create platform-native sizes and distinct hooks. Run policy, brand, and technical QA. Return the approved asset artifacts, but do not upload or publish anything."

"Adapt approved creative [ASSET ID] for [PLATFORMS]. Preserve the core claim, but rewrite copy and resize assets for each platform's format. Return a variant matrix and QA results. Do not publish."

Landing Pages

"Audit [LANDING PAGE URL] for message match, mobile usability, page speed, form friction, trust, policy risk, UTMs, and conversion tracking. Prioritize fixes by expected impact."

"For workspace [WORKSPACE ID], prepare a read-only landing-page draft proposal for [CAMPAIGN]. Audience: [DESCRIPTION]. Offer: [OFFER]. Primary CTA: [CTA]. Return the approved brand-guideline version, exact copy and assets, requested slug, conversion event and destination, agent-generated operation ID, and configuration hash. Confirm no matching page or slug exists. Do not create or publish a page."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In workspace [WORKSPACE ID], I approve creating one unpublished landing-page draft at slug [SLUG] using brand version [VERSION], exact copy/assets [CONTENT VERSION], CTA [CTA], and conversion event [EVENT]. Re-read the workspace, brand, content, event, and slug availability; reject any scope, hash, version, content, event, or create-precondition mismatch. Use the operation ID as an idempotency key where supported, then read back the draft ID, slug, publication state, and conversion configuration. Do not publish. Reconcile an ambiguous result before retrying and do not create a duplicate draft."

"Verify draft landing page [PAGE/SLUG] on mobile and desktop. Test the primary CTA and conversion event, inspect the final destination, and list blockers before publication."

Commerce, Lifecycle, and Events

"Pull Shopify sales, product, and inventory data from [START DATE] through [END DATE]. Identify products with enough margin and stock for paid promotion. Do not create discounts or campaigns."

"In Shopify connection [CONNECTION], store/account [STORE ID], preflight exporting customers matching [EXACT FILTER] to encrypted staged artifact [ARTIFACT NAME] owned by tenant/workspace [TENANT/WORKSPACE ID]. Return the source snapshot/version or cutoff, selected schema and field mapping, expected row count, retention policy, agent-generated operation ID, and configuration hash. Do not export, display customer data, or sync an audience. After the artifact is separately approved and created, propose destinations with exact platform connection and account/advertiser IDs, one new operation ID and hash per destination, immutable artifact ID/version/digest, owning tenant/workspace, row count, mapping, and no-matching-audience result. Each sync requires another approval repeating those values and the exact audience name; re-read and reject mismatches, use idempotency keys where supported, and report native IDs and processing states without displaying PII."

"In Klaviyo connection [CONNECTION], workspace/account [WORKSPACE/ACCOUNT ID], prepare a read-only campaign-draft proposal for segment [SEGMENT ID] at version [SEGMENT VERSION] using template [TEMPLATE ID] at version [TEMPLATE VERSION]. Return the sender identity, subject, preview text, exact content, exclusions, exact current recipient count, agent-generated operation ID, and configuration hash. Do not create a Klaviyo campaign object or send anything."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Klaviyo connection [CONNECTION], workspace/account [WORKSPACE/ACCOUNT ID], I approve creating one unsent campaign draft using segment [SEGMENT ID] at version [SEGMENT VERSION], template [TEMPLATE ID] at version [TEMPLATE VERSION], sender [SENDER], subject [SUBJECT], preview text [PREVIEW], content version [CONTENT VERSION], exclusions [EXCLUSIONS], and expected recipient count [COUNT]. Re-read the scope, segment, template, exclusions, and recipient count immediately; reject any scope, hash, version, content, exclusion, recipient-count, or create-precondition mismatch. Confirm no matching draft exists, use the operation ID as an idempotency key where supported, and read back the draft ID and complete configuration without sending it. If completion is ambiguous, reconcile by scope, operation ID, name, and configuration before retrying; do not create a duplicate draft."

"In Loops connection [CONNECTION], workspace [WORKSPACE ID], prepare a read-only campaign-draft proposal for audience [AUDIENCE ID/VERSION]. Return the sender identity, subject, exact content, exclusions, exact recipient count, expected draft state, agent-generated operation ID, and configuration hash. Do not create a campaign object or send anything."

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In Loops connection [CONNECTION], workspace [WORKSPACE ID], I approve creating one unsent campaign draft for audience [AUDIENCE ID/VERSION] with sender [SENDER], subject [SUBJECT], content version [CONTENT VERSION], exclusions [EXCLUSIONS], and expected recipient count [COUNT]. Re-read every value and confirm no matching draft exists; reject any scope, hash, audience-version, content, exclusion, recipient-count, or create-precondition mismatch. Use the operation ID as an idempotency key where supported and read back the draft ID and full configuration without sending. If completion is ambiguous, reconcile by scope, operation ID, and configuration before retrying; do not create a duplicate draft. Sending requires a new preflight and separate approval."

"Create a Luma event plan for [EVENT] and a retargeting workflow for registrants, attendees, and no-shows. Show the audience mappings and ad-platform destinations before creating or syncing anything."

Demand, Keyword, and Competitive Research

"Find keyword opportunities for [PRODUCT] using Google Ads demand data and Search Console performance. Separate high-intent, comparison, problem-aware, and informational themes. Include volume, competition, and recommended landing page."

"Research currently active ads from [COMPETITORS] across public ad libraries. Summarize repeated hooks, formats, offers, and landing-page patterns. Do not copy protected creative or unsupported claims."

"Build a creative brief from our top-performing ads and competitor whitespace. State which findings come from our performance data, which come from public ads, and which are hypotheses to test."

"Profile the market for [ICP] using buying-intent signals. Return named account segments, signal evidence, estimated audience size, and recommended channels. Do not activate audiences."

Guardrails, Alerts, and Autonomous Management

"Propose guardrails for campaign [ID]: maximum daily budget, target CPA range, maximum percentage budget change, minimum evidence before pausing, and escalation conditions. Do not activate autonomous management."

"For [PLATFORM] connection [CONNECTION], account/advertiser ID [ACCOUNT ID], preflight a daily-spend-greater-than-[DAILY THRESHOLD] [ISO ACCOUNT CURRENCY] alert for campaign IDs [IDS], notifying [EMAIL/SLACK/SMS/WHATSAPP]. Return an agent-generated operation ID and configuration hash, live account currency, human threshold, platform API units, exact campaign set, evaluation schedule, time zone, daily spend window, supported channels, and whether an equivalent alert already exists. Do not create it. Create only after my second approval repeats the operation ID, hash, scope, campaign set, exact alert configuration, account currency, human threshold, and API units. Re-read immediately and reject any scope, hash, currency, units, threshold, or campaign-set mismatch. If this platform cannot evaluate a daily threshold, do not substitute a weekly threshold; return the required daily manual monitoring step instead. Reconcile an ambiguous result before retrying."

"Audit active campaigns for delivery, tracking, budget pacing, and anomalies from [START DATE] through [END DATE]. Separate urgent blockers from optimization opportunities. Do not mutate campaigns."

"Review every change made by the autonomous agent in the last [N] days. Show actor, timestamp, before/after state, reason, result, and whether the intended KPI improved."

Reports, Documents, and Sheets

"Create a client-ready monthly performance report for [DATE RANGE]. Include an executive summary, channel results, qualified outcomes, budget pacing, tests completed, and next decisions. Keep internal tool details out of the report."

"Read the approved media plan from Google Sheet [SHEET]. Validate required fields, account mapping, budgets, dates, and totals. Return row-level errors and do not create campaigns."

"For Google Sheet [SHEET ID], preflight writing verified campaign IDs, current statuses, and validation results to approved columns [COLUMNS] and rows [ROWS]. Return an agent-generated operation ID and configuration hash, expected current cell values, and exact target values. Do not write. Write only after my second approval repeats the operation ID, hash, sheet, ranges, expected values, and target values. Re-read and reject changed cells, do not overwrite planning inputs or formulas, and reconcile an ambiguous result before retrying."

MCP, Jobs, and Usage

"List the current tools available for [PLATFORM OR TASK]. Separate read, create, update, and unsupported operations. Do not infer capability from the platform's public API alone."

"Inspect the exact argument contract for action [ACTION] before running it. Show required fields, accepted values, defaults, and whether the action can create spend. Do not execute it."

"Show my current usage balance and forecast the cost of [PLANNED ACTIONS]. Do not run the actions."

"Check job [JOB ID]. If it timed out or stopped streaming, determine whether the underlying action completed before retrying. For create operations, list recent matching entities first to avoid duplicates."

---

Prompt Templates

Analysis only

"In [PLATFORM] account [ACCOUNT], pull [LEVEL] performance from [START DATE] through [END DATE]. Show [METRICS]. I need to decide [DECISION]. Do not change anything. State any data limitations."

Create a reviewed draft

Message 1 — Preflight:

"In [PLATFORM] connection [CONNECTION], account/advertiser ID [ACCOUNT ID], propose one PAUSED [CAMPAIGN TYPE] campaign named [NAME] with budget [AMOUNT] [ISO ACCOUNT CURRENCY] represented as [PLATFORM API UNITS] [DAILY/LIFETIME], objective [OBJECTIVE], audience [AUDIENCE], destination [CONTENT URL WITH UTMS], and creative [ASSET NAME/ID]. Verify tracking, generate an operation ID and configuration hash for the complete proposed hierarchy, and confirm that no matching campaign exists. Do not create anything."

Message 2 — Approval:

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. Exact preflighted configuration (verbatim, not a summary): [FULL PREFLIGHT CONFIGURATION]. In [PLATFORM] connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve creating one PAUSED [CAMPAIGN TYPE] campaign named [NAME]. Budget: [AMOUNT] [ISO ACCOUNT CURRENCY] represented as [PLATFORM API UNITS] [DAILY/LIFETIME]. Objective: [OBJECTIVE]. Audience: [AUDIENCE]. Destination: [CONTENT URL WITH UTMS]. Creative: [ASSET NAME/ID]. Verify tracking. Immediately re-read the scope, account currency, platform units, and no-matching-campaign precondition; reject any scope, configuration-hash, currency, units, or create-precondition mismatch. Use the operation ID as the idempotency key where supported. Build the complete hierarchy and read back every created ID. Do not enable spend. If completion is ambiguous or partial, reconcile the hierarchy and report every created ID. A resume or cleanup requires a new preflight and separate approval; do not replay or claim rollback."

Propose an optimization

"From [START DATE] through [END DATE], list up to [N] [PLATFORM] entities that meet [RULE]. Show exact IDs, current state, evidence, and proposed change. Do not execute."

Apply an approved change

Message 1 — Preflight:

"In [PLATFORM] connection [CONNECTION], account/advertiser ID [ACCOUNT ID], prepare [ACTION] for IDs [IDS] only. Return an agent-generated operation ID, configuration hash, exact current values, resource version or ETag when available, and exact absolute target values. For any monetary value, return the live ISO account currency, human amount, and platform API units. Do not change anything."

Message 2 — Approval:

"Operation ID [OPERATION ID]. Preflight configuration hash [CONFIGURATION HASH]. In [PLATFORM] connection [CONNECTION], account/advertiser ID [ACCOUNT ID], I approve [ACTION] for IDs [IDS] only, from expected before-state and version [EXPECTED STATE AND VERSION] to exact target state [TARGET STATE]. Approved account currency and platform monetary units: [ISO ACCOUNT CURRENCY / HUMAN AMOUNT / API UNITS, OR NO MONETARY VALUE]. Immediately re-read every target and reject any scope, configuration-hash, currency, units, or before-state mismatch. If NO MONETARY VALUE is declared but any target field is monetary, reject the request. Use conditional updates and the operation ID as the idempotency key where supported. Do not modify any other entity. Read each target back with its before-and-after result. If completion is ambiguous, reconcile before retrying and do not replay a write already applied."

Diagnose a failure

"Diagnose [ERROR OR SYMPTOM] for [PLATFORM/ACCOUNT]. Check account selection, parameters, entity hierarchy, permissions, connection health, and platform response in that order. Do not retry a create operation until you prove that no matching entity was created."