# Ad Creative Library Builder > Ad Creative Library Builder - Skill: `raneqbarber/ad-creative-library-builder` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add raneqbarber/ad-creative-library-builder` - Raw SKILL.md: https://api.skillmd.com/api/skills/raneqbarber/ad-creative-library-builder/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: raneqbarber (https://skillmd.com/u/raneqbarber) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/raneqbarber/ad-creative-library-builder --- # Ad Creative Library Builder Builds a searchable library of every ad you have run, in a spreadsheet, with AI doing the assembly. Nearly every creative intelligence project stalls in the same place: nobody can find the ads. The files are across Drive, the platform export has ad IDs but no thumbnails, the naming convention changed twice, and the person who knew where things were has left. This produces the substrate everything else needs — one row per creative, one lookup per ad ID, and a visual audit that lets a human confirm the whole thing is right. It is deliberately a spreadsheet, because a spreadsheet is the only tool a marketing team will still be using in three years. --- ## 1 · Operating contract - **The library is generated whole, never patched.** All four tabs are replaced together. Asset IDs are sequential, so a partial update silently desynchronises the tabs and nothing errors. - **Never show stale data.** If the source cannot be read, show an error banner. A library displaying yesterday's data as today's is worse than one that is down. - **Never hardcode an ID.** Sheet IDs, folder IDs and tab names live in configuration. A hardcoded ID makes the whole thing single-use. - **Never store a secret in the spreadsheet or the script.** Anything sensitive goes in the platform's properties store. - **Every ad ID remains traceable.** The library reduces the unit of analysis, never the data. - **The audit tab is not optional.** A library nobody can verify is a library nobody should trust. --- ## 2 · When to run this Run at the start of any creative intelligence project, and whenever a new brand or account is onboarded. Run `duplicate-ad-creative-finder` first — the library is built on deduplicated creative IDs, and building it on ad IDs produces a library that repeats the same creative dozens of times. --- ## 3 · Required inputs and preflight ### Must have | Input | Notes | |---|---| | Platform exports | One row per ad, with ad ID, name, campaign, spend | | Creative file links | Resolvable URLs to the actual media | | Deduplication output | Creative inventory and ad-to-creative map | | A spreadsheet | Google Sheets, or any equivalent with a scripting layer | ### Improves the output Trait tags, launch dates, placements, per-ad performance, and caption text. ### Preflight 1. **Deduplication run?** *Halt if not.* Everything here keys on creative ID. 2. **Links resolvable?** Sample and confirm. Report any systematic failure pattern before building. 3. **Write access confirmed?** To the sheet and to the script environment. 4. **Configuration collected?** Sheet ID, tab names, header row count, column positions. All of it into config, none of it into code. --- ## 4 · The procedure ### Step 1 — Build the four tabs | Tab | Rows | Answers | |---|---|---| | **creative_inventory** | One per unique creative | The library. Aspect ratio, duration, slide count, canonical file URL, first and last seen | | **ad_mapper** | One per platform ad | Given an ad ID, which creative is it | | **asset_rollup** | One per creative | Given a creative, how many ads used it and which — sorted by reuse, so the workhorses sit on top | | **creative_audit** | One per ad | The proof. Assigned thumbnail beside that ad's own original file | Four tabs because they answer four different questions, and a single mega-tab answers none of them well. ### Step 2 — Set the header structure Multi-row headers — a title row, a section-group row, then the actual column names — are readable for humans and hostile to scripts. Record the header row count in configuration and have every reader use it. Getting this wrong produces an off-by-one that silently shifts every column, and the data looks plausible. ### Step 3 — Write the backend A read-only script that pulls from the sheet on load. Three rules: - **Read live every time.** No cached copy, no fallback data. Cached data becomes stale data and stale data gets acted on. - **Fail loudly.** If the sheet cannot be read, show a banner saying so. Never fall back to a hardcoded sample — someone will eventually report on it. - **Fixed column positions from config.** Name-matching headers looks robust and breaks the first time someone edits a header. The sheet is the admin panel. There is no separate database and no CMS, which is what makes it maintainable by whoever inherits it. ### Step 4 — Build the audit tab properly One row per ad, showing the assigned creative's thumbnail **beside** that ad's own original file. Any row where the two images differ is a bad match. Add a `Match` column with `representative` or `merged duplicate`. To audit fast, filter to merged duplicates only — those are the only rows where a match can be wrong, since representatives are true by definition. On a 1,006-ad library that is roughly 800 rows to check rather than 1,006. ### Step 5 — Handle media that does not render Cloud storage links frequently will not embed in a spreadsheet cell. Convert share links to direct-view endpoints, handle virus-scan interstitials, and where a file cannot render, show the filename and a link rather than a broken image. A library of broken image icons gets abandoned in a week. ### Step 6 — Write the rebuild instructions The deliverable is not the library, it is the ability to rebuild it. Document: which config values exist and what each controls, the regeneration order, what breaks if tabs are partially updated, and how to add a new brand. ### Step 7 — Verify before handing over - Row counts reconcile across all four tabs - Every ad ID appears exactly once in `ad_mapper` - Every creative in `inventory` appears in `asset_rollup` - Spot-check 20 audit rows visually - Break the source deliberately and confirm the error banner appears That last check is the one people skip, and it is the one that matters — an untested failure path is an untested failure path. --- ## 5 · Computation rules - **All four tabs regenerated together.** - **No hardcoded IDs anywhere.** Config only. - **No cached or fallback data.** - **Fixed column positions from config**, never header-name matching. - **Every ad ID traceable to its creative.** - **Audit tab always generated.** --- ## 6 · Output contract ``` CREATIVE LIBRARY · [brand] · built 2026-09-02 TABS creative_inventory 515 rows unique creatives incl. carousel cards ad_mapper 1,006 rows one per platform ad asset_rollup 202 rows one per creative, sorted by reuse creative_audit 1,006 rows assigned thumbnail vs original RECONCILIATION ad_mapper rows = source ad rows 1,006 = 1,006 ✓ every ad ID appears exactly once ✓ every inventory creative in rollup ✓ audit rows = ad rows ✓ AUDIT representative 200 rows true by definition merged duplicate 806 rows ← check these spot-checked 20 rows 2 mismatches found, threshold tightened MOST REUSED IMG-1043 17 ads across Meta and LinkedIn VID-1012 11 ads CAR-0217 9 ads CONFIG — nothing hardcoded SHEET_ID in script properties TAB names config HEADER_ROWS 3 column map config, fixed positions FAILURE PATH — tested source made unreadable → error banner displayed, no data shown ✓ TO REBUILD 1. run duplicate-ad-creative-finder → creative inventory + ad map 2. regenerate all four tabs together, never individually 3. reconcile row counts 4. spot-check 20 merged-duplicate audit rows 5. break the source, confirm the banner, restore ``` --- ## 7 · Failure and degraded states | Situation | What to do | |---|---| | Deduplication not run | Halt. A library keyed on ad IDs repeats the same creative dozens of times. | | Links not resolvable | Report the pattern. Systematic failure is usually casing or permissions, not missing files. | | Media will not render | Show filename and link. Never a broken image. | | Partial update requested | Refuse. Explain that sequential IDs desynchronise silently. | | No script environment | Build the tabs as static exports and say the library will not stay current. | | Row counts do not reconcile | Halt and report the gap. Do not hand over an unreconciled library. | | Audit finds mismatches | Report them, tighten the deduplication threshold, and rebuild. | --- ## 8 · Worked example Above. Two design decisions determine whether the library is still in use a year later. **No fallback data, ever.** The obvious convenience is to cache a copy so the page still renders when the sheet is unreachable. That convenience is how a team ends up reporting on data from three weeks ago without knowing it. An error banner is annoying for ten minutes; silent stale data is wrong for a quarter. **The audit tab filtered to merged duplicates.** 1,006 rows is not auditable and nobody will do it. 806 rows is still a lot — but it is the *complete* set of rows where a match could be wrong, and knowing that makes a spot-check of 20 statistically meaningful rather than arbitrary. In this run it found two mismatches, which changed the deduplication threshold and improved every downstream analysis. --- ## 9 · Self-check 1. Were all four tabs generated together? 2. Do row counts reconcile across all four? 3. Is every ad ID traceable to exactly one creative? 4. Are all IDs and tab names in config rather than in code? 5. Does the audit tab show assigned thumbnail beside original, with a match column? 6. Was the failure path tested by breaking the source deliberately? 7. Are rebuild instructions written for someone who was not here? --- *MIT licensed.*