agentsclimarketplace

Tiktok creative and catalog optimization

Skill scumunna/programmatic-skills/skills/tiktok-creative-and-catalog-optimization

Agent skills for programmatic trading, analytics, and account operations. DV360 first, multi-DSP and multi-runtime (Claude Code and Codex).

Install
npx -y skills add scumunna/programmatic-skills --skill tiktok-creative-and-catalog-optimization

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Build and optimize TikTok campaigns through creative, catalog, Smart+, pixel and Events API readiness, asset groups, product sets, creative fatigue, audience controls, budget, learning, and reporting. Use when the user asks how to improve TikTok ads, Smart+ Web, catalog ads, TikTok Shop or product catalog performance, creative testing, or creator-style direct-response ads.

SKILL.md

8.5 KB, ~1.8k tokens by cl100k_base, as published. Nobody here has run it

TikTok creative and catalog optimization

Help an operator improve TikTok campaigns by feeding the system better creative, cleaner catalog data, and a stronger conversion signal. TikTok's automation can handle targeting, bidding, and creative rotation, but it cannot invent a clear offer, fix a broken event feed, or make a weak catalog sell. This skill supports the campaign operator. It does not replace review, approval, or spend control.

For server-side event delivery and deduplication, use tiktok-events-api-server-side. For Smart+ campaign setup and the unified builder, use tiktok-smart-plus-and-campaigns.

When to use this skill

  • "My TikTok campaign is not converting. What do I check?"
  • "How do I improve Smart+ Web or Smart+ Sales performance?"
  • "How should I structure TikTok creative tests?"
  • "What should I feed the asset group?"
  • "How do I use TikTok catalog ads or product sets?"
  • "Creative fatigue is killing performance."
  • "How do I brief creator-style direct-response ads?"
  • "How do I read TikTok performance without chasing cheap clicks?"

Boundaries with sibling skills:

  • Smart+ campaign build, objectives, automation modes, and API model: tiktok-smart-plus-and-campaigns.
  • Events API, event dedup, match keys, and web/app events: tiktok-events-api-server-side.
  • Cross-platform creative test design: direct-response-creative-testing.
  • Incrementality: incrementality-and-experimentation.
  • Conversion reconciliation: cross-platform-conversion-reconciliation.

Quick reference

LeverWhat the operator controls
SignalPixel and Events API health, event_id dedup, match keys, conversion objective
CreativeHook, first 2 seconds, UGC or creator style, offer, proof, CTA, asset volume
CatalogProduct feed quality, product sets, image/video quality, price, availability, landing URL
AudienceGeo, language, exclusions, retargeting, customer lists, broad versus constrained setup
BudgetCampaign or ad group budget, learning stability, scale steps
AutomationFull, partial, or manual modules in the unified Smart+ flow
MeasurementCPA, ROAS, CVR, quality, assisted impact, incrementality

Core process

  1. Confirm the objective and optimization event. Sales, lead generation, app promotion, and traffic each optimize differently. A sales campaign with a traffic objective buys cheap visits instead of buyers.
  2. Verify event flow before creative tuning. Confirm pixel and Events API events are firing, deduped, consented, and mapped to the selected objective. A broken signal makes every creative read suspect.
  3. Check catalog health for product-led campaigns. Product ID, title, image, price, availability, URL, category, product set, and TikTok Shop or commerce connection must be correct. Bad feed data creates bad dynamic ads.
  4. Build creative for TikTok context. Lead with a strong hook, show the product or proof quickly, keep the offer obvious, use native-feeling pacing, and make the CTA concrete. Repurposed horizontal TV spots rarely win without edits.
  5. Feed asset groups with meaningful variety. Use multiple hooks, creators, proof types, offers, and formats. Do not upload near-duplicates and expect learning. Label concepts so reporting can be read.
  6. Decide automation module by module. Full Smart+ works when signal and creative volume are strong. Manual controls are appropriate for regulated categories, hard geo, brand safety, or budget-split constraints.
  7. Protect learning. Avoid changing objective, optimization event, budget model, bid target, and creative set daily. Batch edits and let the model stabilize.
  8. Manage fatigue with a pipeline, not a panic refresh. Watch frequency, CTR, thumb-stop signals, CVR, CPA, and creative-level spend concentration. Refresh hooks and creators before fatigue forces a full rebuild.
  9. Read post-click quality. Cheap clicks or high watch time do not matter if the landing page, product page, or checkout fails. Compare click-to-view, add-to-cart, purchase, lead quality, and refund or return signals where available.
  10. Human-gate changes. The skill can recommend new assets, product-set changes, budget steps, and pauses. A person approves live status and budget changes.

Decision rules and thresholds

  • Creative is the main targeting input on TikTok. If creative does not signal the right buyer and use case, broad automation has little to work with.
  • Do not judge a concept on one file. Read concept families across multiple creators and hooks.
  • For catalog ads, feed quality is media quality. Product data, availability, image quality, and URL accuracy decide whether dynamic creative can work.
  • Use retargeting carefully. Retargeting can inflate ROAS by capturing people already close to purchase. Use exclusions and incrementality checks for budget decisions.
  • Scale in steps after learning. A large budget jump changes delivery and can reset stability. Increase gradually, then re-read CPA or ROAS.
  • Add creative before squeezing bids. When performance degrades with rising frequency, fresh creative usually beats a tighter bid.
  • Optimize lead campaigns to quality. Add qualifying form fields or offline quality feedback when cheap leads waste sales time.

Creative testing plan

Test axisExample variantsWhat to read
HookProblem, proof, offer, objectionThumb stop, CTR, CVR, CPA
CreatorCustomer, expert, employee, founderTrust, conversion rate, comments
FormatUGC, demo, testimonial, catalog cardCPA, ROAS, completion, click quality
OfferDiscount, bundle, trial, guaranteeCVR, ROAS, margin, refund risk
Product setBest sellers, high margin, new arrivalsROAS, AOV, availability, stockouts

Catalog QA checklist

  • Product IDs are stable and match event payloads.
  • Titles are clear and not keyword-stuffed.
  • Images or videos are high quality and current.
  • Prices, sale prices, and availability are accurate.
  • Product URLs land on the exact product.
  • Product sets map to campaign intent.
  • Out-of-stock and disapproved items are excluded.
  • Event payload product IDs match catalog IDs for dynamic optimization.
  • TikTok Shop or commerce connection is healthy where used.

Templates and examples

Smart+ Web sales improvement:

Signal: Purchase event firing through pixel and Events API, event_id dedup verified.
Catalog: Best-seller product set, out-of-stock products excluded, product IDs match events.
Creative: 12 assets across 4 hooks, 3 creator styles, same offer.
Optimization: Hold budget and objective for learning, refresh weakest hook family after read.
Decision: Scale only if CPA and ROAS clear target and catalog availability holds.

Creative fatigue response:

Signal: Spend concentrated on 2 assets, frequency up, CTR and CVR down, CPA rising.
Action: Brief 6 new hooks using the winning proof angle, keep offer and landing page stable.
Do not: Change audience, objective, and bid at the same time.

Common pitfalls

  • Launching sales optimization before purchase events and catalog IDs match.
  • Feeding Smart+ one or two weak creative files and blaming the algorithm.
  • Reusing polished TV creative without a TikTok-native hook.
  • Treating watch time as success when CPA or ROAS misses.
  • Leaving out-of-stock products in a product set.
  • Changing objective, budget, and creative together, then losing the test read.
  • Scaling retargeting ROAS without checking incrementality.
  • Letting generated or creator-submitted claims skip brand and legal review.

Sources

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most performance cost skills give in ~1.8k tokens

Counted across 803 of the 1,058 authors here whose files we hold, read 2026-08-07

  • Keep skill files under 500 lines or tokensin 82 of 803, across 16 files
  • Use imperative form in instructionsin 80 of 803, across 9 files
  • Draft assertions while test runs are in progressin 75 of 803, across 9 files
  • Create two to three realistic test promptsin 74 of 803, across 9 files
  • Write skill descriptions to be pushyin 72 of 803, across 7 files
  • Save test cases to evals JSONin 72 of 803, across 6 files
  • Ask questions about edge cases and input formatsin 72 of 803, across 7 files
  • Save timing data immediately when runs completein 70 of 803, across 5 files
  • Include all trigger conditions in the skill descriptionin 69 of 803, across 3 files
  • Launch all test runs in a single turn or simultaneouslyin 69 of 803, across 3 files
  • Capture intent before writing a skillin 67 of 803, across 1 file
  • Import directly instead of barrel filesin 52 of 803, across 15 files

Said here and by no other author read

  • confirm the campaign objective and optimization event
  • verify pixel and events api health
  • check catalog health and product feed quality
  • build creative using strong hooks and clear offers
  • feed asset groups with varied creative concepts
  • protect learning by batching edits

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.