Back to search
ads-creative-lab
MarketplaceWorks with Claude CodeGitHub
A paid-social creative loop: concepts, safe-zone checks, a paused launch and a read on results.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/mavrin-xv--ads-creative-lab)HTML
<a href="/listings/mavrin-xv--ads-creative-lab"><img src="/api/listings/mavrin-xv--ads-creative-lab/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Every script runs offline on Acme Coffee, an invented small roaster.
python3 skills/creative-brainstorm/scripts/score_concepts.py examples/acme-coffee/concepts.json
## Concept scorecard (7 pass, 0 fail the gates) | Rank | ID | Format | Motive | Score | Novelty | |---|---|---|---|---|---| | 1 | AC-C01 | static | save time | 86.0 | new | | 2 | AC-C02 | ugc | trust | 86.0 | new | ...
python3 skills/ad-results-readout/scripts/readout.py examples/acme-coffee/ads-export.csv --target-cpa 25 --plan examples/acme-coffee/cells.txt --orders examples/acme-coffee/orders.csv
| Ad | Spend | Share | Purchases | CPA | CTR % | Verdict | |---|---|---|---|---|---|---| | AC-C02-ugc-9x16 | 118.56 | 28.8% | 8 | 14.82 | 1.58 | **winner** | | AC-C01-static-9x16 | 63.06 | 15.3% | 0 | - | 1.39 | **kill** | ... - Unmapped spend 71.93 (17%): AC-OLD-promo, AC-C05-carousel-4x5. Name every ad with its cell ID before launch. - Denominator 2026-03-23 to 2026-03-29: shop recorded 79 orders, the ad account claims 13 purchases (16.5% of all orders).
Every other script's output is in [examples/acme-coffee/outputs/](examples/acme-coffee/outputs/), and the commands to rerun them are in [examples/README.md](examples/README.md).
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 15 lines
- {
- "name": "ads-creative-lab",
- "plugins": [
- {
- "name": "ads-creative-lab",
- "source": "./",
- "description": "A paid-social creative loop for small brands as agent skills: competitor scan, scored and deduplicated concepts, 9:16 and 4:5 statics checked against safe zones, product photo prompts, a launch built paused, Meta CAPI dedup, a rules-first readout and a pause test.",
- "version": "0.1.0"
- }
- ],
- "owner": {
- "name": "Artem Mavrin",
- "url": "https://github.com/mavrin-xv"
- }
- }
README.md · README · 150 lines
- 
- # ads-creative-lab
- **A paid-social creative loop for small brands, run by your AI agent: competitor scan, defended concepts, 9:16 and 4:5 statics checked against safe zones, product photo prompts that keep the product real, a launch built PAUSED, server-side tracking that dedupes, a readout with rules written first, and a pause test for "do ads work at all".**
- Small brands lose ad money in the gaps: ten hooks that are one idea, a headline under the reels caption, an ad set live by accident, purchases counted twice, a winner called on day two.
- [](LICENSE)
- 
- 
- 
- ```bash
- git clone https://github.com/mavrin-xv/ads-creative-lab && cd ads-creative-lab
- ./install.sh # or: ./install.sh --project
- ```
- <!-- install-notes:start -->
- Other ways in: as a Claude Code plugin, run `/plugin marketplace add mavrin-xv/ads-creative-lab` and then `/plugin install ads-creative-lab@ads-creative-lab`. For Codex or another agent that reads `SKILL.md`, copy the folders under `skills/` into its skills directory. On Windows, run `install.sh` from Git Bash or WSL and type `py -3` where these docs say `python3`.
- <!-- install-notes:end -->
- Then tell your agent: `make a new batch of ads for my shop` or `read this ads export` or `are my ads bringing sales at all?`
- ---
- ## What it solves
- | Pain | What the pack does |
- |---|---|
- | Ten ad ideas that are the same idea | `score_concepts.py` applies hard gates, a weighted score and a novelty test on motive, belief, mechanism and proof, against the batch and every concept you shipped before |
- | Ideas built on numbers nobody checked | A claims ledger: every fact is `verified`, `needs_verification` or `do_not_use` before it reaches an image |
- | Text under the reels buttons, a font that fell back | `render_pair.py` renders 9:16 and 4:5 from one template and checks the font loaded; `safe_zone_check.py` measures every box against placement bands, crop lines and each other |
- | AI images that "improve" the product | `shot_prompt.py` writes prompts with a scale sentence, a do-not-invent list and "keep the product exactly as photographed" |
- | An ad set went live before anyone looked | The launch checklist builds everything PAUSED; activation needs the owner's yes to a named list |
- | A test budget that can never reach a verdict | `test_budget.py`: cells x 2 x target CPA, days needed, kill threshold and live ads per budget |
- | Purchases counted twice, or not at all | `capi_event.py` builds hashed server events, lints them, sends to Test Events (dry run by default) and finds pairs that will not dedupe |
- | "This one looks promising" after two days | `readout.py` gives each ad one verdict from rules on the test card, and flags unmapped spend, fatigue and the shop vs platform gap |
- | "Would we get these customers anyway?" | `event_log.py` runs a pause test on a call log or an order log against the normal weekly swing |
- ## What's inside
- | Skill | What it does | What to say |
- |---|---|---|
- | `ad-creative-loop` | The spine: nine stages with a gate each, post-launch rules, round close | "run ads for my brand", "make a new batch" |
- | `ad-library-scan` | Logs competitor ads in one CSV, finds the longest runners and your gaps | "what are competitors running", "look at the ad library" |
- | `creative-brainstorm` | 12 exercises, 30 techniques, 20 angles, 45 hook frames, claims ledger, scoring and dedupe | "ad ideas", "new angles for our winner" |
- | `ad-variants` | 9:16 and 4:5 statics from one template, font probe, safe zones, overlaps, contact sheet; video rules | "make the ad images", "check safe zones" |
- | `product-photoshoot-prompts` | Ten shot modes, model-agnostic prompts that keep the product real | "product photos for ads", "lifestyle shot prompts" |
- | `ad-launch-checklist` | Preflight, budget maths, structure, naming and UTMs, PAUSED build, activation on yes | "launch the ads", "how much budget for the test" |
- | `meta-capi` | Server events with correct hashing, lint, test send, dedup check, browser checks | "CAPI", "purchases counted twice" |
- | `ad-results-readout` | One verdict per ad, account checks, loser diagnosis | "which ad won", "read the results" |
- | `ads-incrementality-check` | Pause test on calls or orders, four verdicts, money check | "are my ads working at all", "pause test" |
- ## Quick start
- Every script runs offline on Acme Coffee, an invented small roaster.
- ```bash
- python3 skills/creative-brainstorm/scripts/score_concepts.py examples/acme-coffee/concepts.json
- ```
- ```
- ## Concept scorecard (7 pass, 0 fail the gates)
- | Rank | ID | Format | Motive | Score | Novelty |
- |---|---|---|---|---|---|
- | 1 | AC-C01 | static | save time | 86.0 | new |
- | 2 | AC-C02 | ugc | trust | 86.0 | new |
- ...
- ```
- ```bash
- python3 skills/ad-results-readout/scripts/readout.py examples/acme-coffee/ads-export.csv --target-cpa 25 --plan examples/acme-coffee/cells.txt --orders examples/acme-coffee/orders.csv
- ```
- ```
- | Ad | Spend | Share | Purchases | CPA | CTR % | Verdict |
- |---|---|---|---|---|---|---|
- | AC-C02-ugc-9x16 | 118.56 | 28.8% | 8 | 14.82 | 1.58 | **winner** |
- | AC-C01-static-9x16 | 63.06 | 15.3% | 0 | - | 1.39 | **kill** |
- ...
- - Unmapped spend 71.93 (17%): AC-OLD-promo, AC-C05-carousel-4x5. Name every ad with its cell ID before launch.
- - Denominator 2026-03-23 to 2026-03-29: shop recorded 79 orders, the ad account claims 13 purchases (16.5% of all orders).
- ```
- Every other script's output is in [examples/acme-coffee/outputs/](examples/acme-coffee/outputs/), and the commands to rerun them are in [examples/README.md](examples/README.md).
- ## How it works
- ```
- brand-ads-playbook.md (tone, claims ledger, what we learned)
- |
- [1] brief -> [2] library scan -> [3] concepts -> [4] statics + video -> [5] owner approves
- gaps? gates pass? safe zones exit 0?
- |
- [6] launch, all PAUSED -> [7] CAPI test + dedup -> hands off 3 days -> [8] readout -> next round
- preflight green events seen once verdict per cell
- |
- [9] pause test when "do ads work at all?" <-+
- ```
- Each stage writes a file the next one reads: the brief, `library-log.csv`, `concepts.json` and the concept bank, renders with manifests, the test card, the export. Results flow back into the playbook, so round two starts from what round one learned.
- ## Configure
- | Setting | Where | Default |
- |---|---|---|
- | Target CPA, kill multiple | `--target-cpa`, `--kill-at` on `readout.py` and `test_budget.py` | kill at 2.5 x CPA with 0 purchases |
- | Score weights and novelty keys | top of `score_concepts.py` | relevance 20, proof 20, clarity 15, distinct 15, brand 10, hook 10, feasible 10 |
- | Safe zones | `safe_zone_check.py --preset` or `--zone` | `reels-9x16` |
- | Graph API version | env `META_GRAPH_VERSION` | v25.0 (check `meta-facts.md`) |
- | CAPI credentials | env `META_PIXEL_ID`, `META_CAPI_TOKEN` | unset: `send` stays a dry run |
- | Your brand | `templates/brand-ads-playbook.md` | blank template |
- Install without the script: copy the folders in `skills/` to `~/.claude/skills/` (or your agent's skills folder).
- ## Honest limits
- - Concept scores rank ideas for testing. They do not predict cost per purchase.
- - Safe-zone percentages come from third-party guides, not the platform. Platform interfaces change; check previews.
- - Platform facts in `meta-facts.md` age. Each has a check date and a 90-day TTL.
- - The ad library shows no spend or results. A long run means the ad survives, not that it converts.
- - A pause test needs enough daily events and a quiet calendar; with few events the honest answer is "can't tell".
- - Meta only. Other platforms need their own zones and structure.
- - `render_pair.py` needs Playwright with Chromium; `contact_sheet.py` and `--overlay` need Pillow. The rest is standard library.
- - All example businesses, rows and numbers are invented. Not tested against a live ad account: `send --live` and the API reads were not run. Tested on Windows with Git Bash and Python 3.12.
- ## Credits
- - Based on / inspired by https://github.com/boringmarketer/boring-marketing-skills (MIT): campaigns created paused, hard-rules blocks, dated platform facts with a TTL, prompt-only mode without a token, crop maths, the pause-test method, test budget and live-ads-per-budget rules, cell-ID naming, batch diversity, the denominator check
- - Based on / inspired by https://github.com/coreyhaines31/marketingskills (MIT): segment, motive, angle and format as separate axes; evidence tiers
- - Based on / inspired by https://github.com/AgriciDaniel/claude-ads (MIT): read-only by default with a gate before any change; "unknown" as an honest cold-start answer
- - Based on / inspired by Meta developer docs on the Conversions API: https://developers.facebook.com/docs/marketing-api/conversions-api/deduplicate-pixel-and-server-events , https://developers.facebook.com/docs/marketing-api/conversions-api/parameters/customer-information-parameters , https://developers.facebook.com/docs/marketing-api/conversions-api/using-the-api , https://developers.facebook.com/docs/marketing-api/conversions-api/best-practices (docs; facts restated, no text copied)
- - Based on / inspired by Meta Business Help on event match quality and ad review: https://www.facebook.com/business/help/765081237991954 , https://www.facebook.com/business/help/980593475366490 (docs)
- - Shot mode taxonomy inspired by product-photoshoot workflows in commercial image tools such as Higgsfield (idea only, no text copied; license unknown)
- ## Contributing
- See [CONTRIBUTING.md](CONTRIBUTING.md). Notes for coding agents are in [AGENTS.md](AGENTS.md).
- <!-- author:start -->
- ## Author
- Built and maintained by [Artem Mavrin](https://github.com/mavrin-xv). This repo is one of a set of agent skill packs for marketing and operations work. Each one installs on its own; the full list is in [agent-skill-packs](https://github.com/mavrin-xv/agent-skill-packs). If a pack saved you an afternoon, a star helps other people find it.
- <!-- author:end -->
- ## License
- MIT. See [LICENSE](LICENSE).
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.