Communitygithub.com

mardab96/tiktok-ads-claude-skills

Builds a staged TikTok budget plan with entry gates, one change per observation window, stop rules, and a recorded rollback, using T04, T05, T06, and T10 from the shared register. Use when someone types "can we scale this TikTok campaign", "double the budget on the winner", "how fast can I raise the daily budget", "scale before the sale", or "plan the budget increase for next week".

Qu'est-ce que tiktok-ads-claude-skills ?

tiktok-ads-claude-skills is a Claude Code agent skill that builds a staged TikTok budget plan with entry gates, one change per observation window, stop rules, and a recorded rollback, using T04, T05, T06, and T10 from the shared register. Use when someone types "can we scale this TikTok campaign", "double the budget on the winner", "how fast can I raise the daily budget", "scale before the sale", or "plan the budget increase for next week".

Compatible avec✓Claude Code~Codex CLI~Cursor
npx skills add https://github.com/mardab96/tiktok-ads-claude-skills/tree/HEAD/budget-scaling-planner-tiktok-ads

Demander à votre IA préférée

Ouvre une nouvelle conversation avec cette compétence d'agent déjà préchargée.

Documentation

Budget Scaling Planner

Use this skill when

A campaign or ad group looks healthy and someone wants more spend. This skill decides whether the account has earned a raise, sets the step size and spacing, writes the stop rules before any money moves, and records the rollback. It never executes a change.

Common requests:

  • "This ad group hit target CPA. Can we double budget?"
  • "What is a safe step size on TikTok?"
  • "Plan the raise for the sale on the 15th."
  • "Budget is capped every day by noon. What now?"
  • "We scaled last week and CPA went up. Roll back or hold?"

Do not use this skill for these jobs:

  • Choosing or changing a bid strategy: bid-strategy-check-tiktok-ads.
  • Edits while learning is in progress: learning-phase-guard-tiktok-ads.
  • CPA already up with no clear cause: cpa-spike-diagnosis-tiktok-ads.
  • Shop profit under GMV Max: gmv-max-profit-check-tiktok-ads.
  • Exhausted creative: creative-fatigue-detector-tiktok-ads.

Required input

Work from pasted exports and screenshots. Claude's TikTok for Business MCP connector is optional.

Required

  • Daily export at ad group or campaign level with columns date, campaign, ad group, budget, spend, impressions, clicks, reported conversions, CPA.
  • Target CPA and where it came from (margin, lead value, owner statement).
  • Bid strategy label and learning status, screenshotted as labelled in your Ads Manager.
  • Dated change log for the last 14 days.
  • Events Manager diagnostics capture.
  • Capacity statement from the owner: stock, fulfilment, sales or call handling, landing page limits.
  • Conversion lag in days and attribution setting.

Optional

  • Independent order or lead report for the same dates.
  • Creative inventory with launch dates.
  • Promo calendar with fixed dates.
  • Budget type (daily or lifetime) at each level.

Export route: ad group table with date breakdown and the budget column visible. See ../tiktok-ads-shared/references/getting-your-data.md.

Before analysis

  1. Confirm the budget type per level. TikTok's About Budget page (July 2026) says a live campaign or ad group cannot switch between daily and lifetime, and a campaign budget should not sit below its ad group budget.
  2. Identify the entity to raise and the one metric that proves it is working, normally CPA on the primary event.
  3. Read the current per-adjustment ceiling and spacing on TikTok's About Budget page. Record the figure and the date as [platform guidance]. T06 is the planned size in the register and stays at or under any ceiling TikTok shows.
  4. Exclude days inside the conversion lag from every CPA you use.
  5. Count mature conversions per window. T04 needs at least 10 per window to judge CPA movement.

Analysis workflow

  1. Check the gates in order (see Decision rules): tracking clean (T10), learning cleared (T05), CPA at or under target on mature rows (T04 floor), strategy label confirmed, capacity confirmed. Any failed gate ends the plan with its verdict.
  2. Read pacing. Compare spend to budget across mature days. An entity that does not spend its budget has a delivery constraint, so a bigger budget changes nothing; route to bid-strategy-check-tiktok-ads.
  3. Set the baseline: mature CPA, results per day, and CPM for the window before the first step.
  4. Draft the ladder. Step size follows T06: 20% per step after learning, one budget action per observation window. An observation window is at least the conversion lag and must hold at least 10 conversions to judge CPA. Compounding applies, so three steps from 100 give 120, 144, and 172.80.
  5. Name the owner of each step and the date it may be applied. Put promo dates in the calendar so no step lands inside a spike window.
  6. Write stop rules before the first step: T04 against the pre-step baseline, T10 pattern, learning re-entry, capacity breach, pacing failure.
  7. Write the rollback per step: the exact prior budget, who restores it, and the trigger. State that restoring the budget does not promise the prior CPA returns.
  8. If TikTok guidance suggests it for the label in use, consider adding a stabiliser. TikTok says Target Cost per Result can help control volatility while budget rises; confirm the strategy with bid-strategy-check-tiktok-ads first, since a strategy switch is its own edit.
  9. Output the plan and the approval request.

Decision rules

Gates:

GatePass conditionIf failedVerdict
TrackingNo T10 pattern, diagnostics cleanStop; fix events firsthold
LearningPast about 25 results or 7 days, status cleared on screen (T05)Wait; route to guardhold
EfficiencyMature CPA at or under target with at least 10 conversions (T04 floor)No raisehold or needs_data
StrategyLabel confirmed by screenshotAsk for itneeds_data
CapacityOwner confirms stock, handling, page can carry moreDo not planhold
Independent checkIndependent report agrees directionally (T07)Explain gap firstneeds_data

Step and stop rules:

ConditionVerdictNext step
All gates pass, step window clearapproval_neededApply one step of 20% (T06) and open the next window
Step window CPA at least 30% above pre-step baseline, both windows at least 10 conversions (T04)likely stopRoll back to the prior budget, then diagnose
Step window has under 10 conversionsneeds_dataExtend the window, apply no second step
T10 pattern appears after a stepholdpixel-events-api-check-tiktok-ads, freeze ladder
Ad group re-enters learning (status on screen)holdlearning-phase-guard-tiktok-ads, freeze ladder
Spend stays well under budget across mature dayspossible bid constraintbid-strategy-check-tiktok-ads, no budget raise
Planned step above the ceiling TikTok showsholdReduce to the ceiling and extend the ladder
Client asks to double in one actionholdOffer staged steps; never skip windows
Campaign budget below ad group budgetholdCorrect the pair before any raise
Creative frequency or T01 signals worseningpossibleAdd creative supply before the next step
Independent report diverges from reported conversions (T07)needs_dataattribution-gap-check-tiktok-ads

TikTok's Maximum Results guidance is that a higher budget often raises cost per result after learning. Plan for CPA to move, and decide the acceptable ceiling before step one.

Output format

Fill this template.

## Scaling plan: [campaign / ad group], [currency], [timezone]
Strategy label: [as labelled] | Learning: [status] | Lag: [days] | Target CPA: [value, source]
TikTok ceiling read: [figure, page, date] | T06 step used: [20%]

### Gate check
| Gate | Evidence | Pass or fail | Verdict |
|---|---|---|---|

### Baseline (mature window, [dates])
Spend/day: [..] | Conversions: [..] | CPA: [..] | CPM: [..] | Budget used: [..]

### Ladder
| Step | Earliest date | Budget from | Budget to | Observe until | Stop rule | Rollback value | Owner |
|---|---|---|---|---|---|---|---|

### Findings
| Finding | Evidence | Verdict | Severity | Confidence | Business impact | Next step |
|---|---|---|---|---|---|---|

### Stop and rollback rules
[each rule, its trigger, and the exact restore action]

### Data gaps
[missing diagnostics, capacity, independent report, and the verdict each would firm up]

Practical example

Illustrative numbers only. Maximum Results ad group, daily budget 100 USD, target CPA 40 USD, lag 7 days. Mature 14 day window: 52 reported conversions, CPA 38 USD, budget fully spent on most days. Learning status cleared on screen. Events diagnostics clean. Owner confirms stock and handling can carry roughly double volume. Independent orders sit within a few percent of reported.

Gates all pass, so the plan is approval_needed. Client asks to double to 200 USD at once. That is hold for a single action; stage it instead.

StepBudget fromBudget toObserve
1100.00120.007 days, at least 10 conversions
2120.00144.007 days, at least 10 conversions
3144.00172.807 days, at least 10 conversions

Stop rule: CPA in an observation window at or above 30% over the 38 USD baseline, which is 49.40 USD, rolls the budget back to the prior step (T04). Any T10 pattern or a learning re-entry freezes the ladder. Rollback for step 2 restores 120.00. If a step window shows only 7 conversions, the answer is needs_data and the window extends before step 3.

Common failure modes

  • Applying step 2 before the first window closes and holds enough conversions.
  • Scaling on a window that still sits inside the conversion lag.
  • Raising budget on an entity that does not spend its current one. Delivery is constrained, so look at the bid.
  • Raising budget and bid together, then having no clean read.
  • Scaling through a sale or holiday spike and reading the spike as the effect.
  • Putting a campaign budget below its ad group budget.
  • Trusting in-platform CPA when the independent report disagrees.
  • Scaling a Smart+ campaign without reading its own run guidance, which carries separate limits.
  • Treating rollback as a guarantee. Restoring the budget does not return the prior audience pool or CPA.
  • Letting creative supply lag the budget until frequency climbs.

Route to next

Timing against learning: learning-phase-guard-tiktok-ads. Strategy mismatch: bid-strategy-check-tiktok-ads. Event doubts: pixel-events-api-check-tiktok-ads. CPA rising: cpa-spike-diagnosis-tiktok-ads. Creative: creative-fatigue-detector-tiktok-ads, winning-ad-remix-tiktok-ads. Automated campaign: smart-plus-audit-tiktok-ads. Client update: weekly-client-report-tiktok-ads.

Guardrails

  • Do not make a live account, budget, bid, placement, tracking, Shop, page, or creative change.
  • Do not plan a raise while any gate is failed or unverified.
  • Do not stage more than one budget action per observation window.
  • Do not exceed the per-adjustment ceiling shown on TikTok's budget page.
  • Do not start a stop rule clock inside the conversion lag.
  • Do not promise that CPA, volume, or return will hold at a higher budget.
  • Do not present platform-reported conversions or GMV as profit without the supplied cost stack.
  • Do not hide missing dates, lag, attribution settings, or conversion definitions.
  • Do not claim a result before a controlled measurement exists.

Individual skills in this repo

This repo contains 18 individual skills — each has its own dedicated page.

mardab96/tiktok-ads-claude-skills

Runs an ordered, layer-by-layer triage of an unfamiliar TikTok Ads account (tracking, structure, learning, creative, bidding, landing page) and hands the first blocker to one named sibling skill. Use when the user inherits a TikTok account, onboards a client, asks where to start, says TikTok results dropped with no known cause, asks whether an account is ready to scale, or says "audit this TikTok account" without naming a problem.

mardab96/tiktok-ads-claude-skills

Reconciles TikTok Ads Manager conversions with GA4, Shopify, CRM, or app analytics using a baseline-ratio method, separating click-through from view-through credit, attribution window and timezone differences, and real tracking drift (T07). Use when a TikTok media buyer or client says "TikTok says 200 sales, Shopify says 90", "GA4 does not show our TikTok conversions", "TikTok ROAS looks too good", "the gap got bigger this month", or "which number do we report to the client".

mardab96/tiktok-ads-claude-skills

Checks whether a TikTok ad group's bidding strategy fits its goal, budget, and conversion volume using the labels TikTok currently lists (Maximum Results, Target Cost per Result, value options as shown in the account) and asks for a screenshot before mapping any legacy name. Use when someone types "Maximum Results or Target Cost per Result", "Cost Cap vs Lowest Cost on TikTok", "our target CPA is too low and delivery died", "bid strategy check", or "should we use hybrid bidding".

mardab96/tiktok-ads-claude-skills

Strips personal data from TikTok ad comments, clusters them into objections, questions, proof requests, and noise with counts, links each cluster to product facts, and drafts reply options and hook ideas for approval. Use when the user pastes TikTok comments and asks "what are people objecting to", "mine the comments for hooks", "draft replies to these comments", "what do viewers keep asking", or wants an FAQ and creative briefs built from comment themes.

mardab96/tiktok-ads-claude-skills

Diagnoses a TikTok CPA jump by splitting it into CPM, CTR, and CVR movement between a baseline and a current window with the cpa_decompose.py helper, then checking tracking, changes, mix, creative, and landing page in that order. Use when someone types "TikTok CPA doubled", "cost per purchase is up 40%", "why did CPA spike", "CVR dropped but clicks are fine", or "the client wants to know what broke".

mardab96/tiktok-ads-claude-skills

Detects whether a TikTok ad is tiring by comparing the last seven distinct dates with the preceding seven for frequency and CTR, using the bundled fatigue_scan.py script and a 14-date rule, then builds a control-versus-refresh queue. Use when the user asks "is this TikTok ad fatigued", says CTR is dropping while frequency climbs, wants to know when to refresh TikTok creative, or pastes a daily ad export and asks which ads to replace.

mardab96/tiktok-ads-claude-skills

Turns TikTok Shop and GMV Max reporting into a per-product cost-stack ledger with refunds, affiliate commissions, seller-funded coupons, referral fees, ad spend, COGS, and seller-borne shipping, using gmv_profit.py to flag any incomplete cost stack (T09). Use when a TikTok Shop seller or media buyer says "GMV Max ROI looks great but we are not making money", "is this product profitable on TikTok Shop", "check our TikTok Shop margins", "affiliate commissions are eating us", or before raising a GMV Max budget or ROI target.

mardab96/tiktok-ads-claude-skills

Scores the opening seconds of TikTok ads by computing 2-second and 6-second view rates and a thumb-stop proxy from ad-level export columns, comparing them with like-for-like ads, and scoring the script opening on a five-point rubric. Use when the user asks "which TikTok hook is weak", "what's my hook rate", "why do people scroll past my ad", wants a hook test plan, or pastes a transcript of the first seconds next to video-view columns.

mardab96/tiktok-ads-claude-skills

Sequences planned TikTok edits around the learning phase, separating urgent repairs from optimisation, naming which edits TikTok itself labels as affecting learning, and ordering one observable action at a time with rollback. Use when someone types "can I edit this ad group during learning", "will this reset the learning phase", "campaign is in learning, what can we change", "sequence these changes", or "the client wants five edits this week".

mardab96/tiktok-ads-claude-skills

Checks TikTok Pixel and Events API health from Events Manager screenshots and an Ads Manager export, covering event match signals, event_id deduplication between browser and server events, parameter completeness on value events, and primary-event drops against stable clicks (T10). Use when a TikTok media buyer says "purchases dropped but clicks are flat", "check our TikTok pixel", "is Events API deduplicating", "Events Manager shows a diagnostic warning", "TikTok value does not match our orders", or before any bid, budget, or scale call that depends on TikTok conversion data.

mardab96/tiktok-ads-claude-skills

Reviews TikTok Ads delivery split between the TikTok placement and Pangle or TikTok Ad Network inventory (as labelled in your Ads Manager), testing CPA and downstream quality per placement against T08 and keeping brand safety and suitability as a separate question. Use when a TikTok media buyer says "should we turn off Pangle", "where is our TikTok spend actually going", "automatic placement is eating budget", "network traffic converts badly", "is our TikTok ad showing next to unsafe content", or when a client asks for a placement and brand safety readout.

mardab96/tiktok-ads-claude-skills

Reviews TikTok Search Ads search terms and keywords and decides what to add as a keyword, add as a negative, or watch, with match handling as labelled in the account and a lag-aware waste floor. Use when someone types "review our TikTok search terms", "negative keyword list for TikTok Search Ads", "which queries should become keywords", "broad match is bringing junk", or "Search Ads spend with zero conversions".

mardab96/tiktok-ads-claude-skills

Audits a TikTok Smart+ campaign by mapping what Smart+ automates against what the media buyer still controls (creative supply, audience exclusions, conversion event, budget, bidding), then ranks the risks. Use when someone types "audit our Smart+ campaign", "what does Smart+ actually control", "Smart+ vs manual", "Smart+ spend is up but results are flat", or "can we trust Smart+ with this budget".

mardab96/tiktok-ads-claude-skills

Compares TikTok Spark Ads with brand-handle ads like-for-like by matching objective, optimization event, audience, offer, placements, bidding label, dates, and concept before any metric is read, then records authorization status and designs a matched test when no clean pair exists. Use when the user asks "do Spark ads perform better than our brand ads", "Spark vs non-Spark", "should we boost creator posts", "why is the creator ad cheaper", or pastes ad-level data with a format column and wants a verdict.

mardab96/tiktok-ads-claude-skills

Writes a complete creator-facing UGC brief for TikTok ads, with viewer situation, one message, three hook options, scene structure, claims allowed and forbidden, do and don't lists, disclosure, usage rights, and a Spark authorization note, plus an internal approval checklist. Use when the user says "write a UGC brief for TikTok", "brief a creator", "I need a creator script with approved claims", "what rights do I need to run this creator video as an ad", or has product facts and needs something a creator can film from.

mardab96/tiktok-ads-claude-skills

Finds mature TikTok campaigns, ad groups, and ads that spent past a CPA-based floor with zero reported conversions, after excluding rows still inside the conversion lag, using the wasted_spend.py helper. Use when someone types "where is the wasted spend on TikTok", "which ad groups burn budget with no conversions", "what can we cut before we scale", "zero conversion ads", or "clean this account before the client call".

mardab96/tiktok-ads-claude-skills

Writes the weekly TikTok Ads client report as a plain-language narrative, a scorecard against the client goal, and next-week decisions with owners and approvals, after an integrity check on tracking, lag, and learning status (T04, T05, T07, T10). Use when an agency or freelance TikTok media buyer says "write the weekly TikTok report", "client update is due", "explain this week's TikTok results to the client", "what do we tell the client about CPA going up", or "set up a repeatable TikTok report format".

mardab96/tiktok-ads-claude-skills

Turns one proven TikTok ad into N distinct angle scripts, each with hook, body beats, CTA, on-screen text, and a shot list, after deconstructing why the winner worked and checking every claim against product facts and rights. Use when the user says "remix my winning TikTok ad", "give me new angles on this winner", "I need five scripts like this one", "my best ad is wearing out, what next", or pastes a winning transcript and asks for variations.

Skills associés