Claude Ads Skills That Actually Execute (Google, Meta, LinkedIn)

1. Wasted Spend Hunter (Google Ads)

Save as ~/.claude/skills/wasted-spend-hunter/SKILL.md

---
name: wasted-spend-hunter
description: Find Google Ads search terms that spend with zero conversions,
  then block them with negative keywords through the PaidSync MCP after a
  dry_run preview and an explicit confirm.
---

# Wasted Spend Hunter (Google Ads)

Audit wasted spend in the connected Google Ads account and end the run
with negatives applied, not a list for later.

## Read
1. Run get_wasted_spend_report for the last 30 days.
2. Run get_search_terms_report sorted by cost, same window.
3. Flag every search term with zero conversions and spend above $25.
   Treat $25 as a default. Ask once per account for a different floor.
4. Check existing negative keywords and shared lists first. Never
   propose a term that is already blocked.

## Propose
Show one table: search term, triggering keyword, campaign, spend,
clicks, conversions, proposed match type. Exact match for one-off
junk. Phrase match only when three or more flagged terms share a
root. Close with the total 30-day waste in dollars.

## Execute, gated
1. Call add_negative_keywords with dry_run: true. Show the planned
   change exactly as returned.
2. Ask "Apply these negatives?" and wait. Silence means no.
3. On confirm, run the same call without dry_run. Report what was
   applied and the projected monthly saving.

## Never
- Never block a term that converted in the last 90 days.
- Never apply more than 50 negatives in one confirmed batch.

2. Budget Pacing Checker (cross-platform)

Save as ~/.claude/skills/budget-pacing-checker/SKILL.md

---
name: budget-pacing-checker
description: Check month-to-date pacing on every connected ad platform,
  then correct off-pace budgets through the PaidSync MCP after a dry_run
  preview and an explicit confirm.
---

# Budget Pacing Checker (cross-platform)

## Read
1. Run get_campaign_performance month-to-date for each connected
   platform. Note monthly budget, spend so far, days elapsed, days
   remaining.
2. Project end-of-month spend at the current daily run rate.
3. Flag campaigns pacing under 85 percent or over 115 percent of
   target.

## Propose
One table across platforms: campaign, platform, budget, projected
spend, pace, proposed new daily budget that lands the month on
target. Sort by dollar gap, largest first.

## Execute, gated
Platform write tools:
- Google Ads: update_campaign_budget
- Meta: update_fb_campaign
- LinkedIn: update_linkedin_campaign
- TikTok: update_tiktok_campaign_budget
- Microsoft: update_ms_campaign_budget
For each flagged campaign, run the write with dry_run: true, show
the before and after, then wait for my confirm per platform batch.
Apply only after I confirm.

## Never
- Never change a budget by more than 25 percent in one run. Flag
  bigger gaps for a human decision instead.
- Never create campaigns. This skill only adjusts what exists.

3. Negative Keyword Triage (Google Ads)

Save as ~/.claude/skills/negative-keyword-triage/SKILL.md

---
name: negative-keyword-triage
description: Weekly Google Ads search term triage. Sort new terms into
  block, watch, and harvest, then apply the block list through the
  PaidSync MCP with dry_run and confirm.
---

# Negative Keyword Triage (Google Ads)

Run weekly. Keeps search term hygiene continuous instead of
quarterly.

## Read
1. Run get_search_terms_report for the last 7 days.
2. Run get_keyword_performance for context on the ad groups involved.

## Triage every new term into three buckets
- Block: intent mismatch. Job hunting, DIY, free-seekers, courses,
  competitor brands we never bid on, unrelated queries.
- Watch: ambiguous terms with spend under the account's CPA. Keep a
  running list, revisit next run.
- Harvest: converting terms not yet added as keywords. Propose them
  as exact match additions via add_keywords.

## Execute, gated
1. Shared junk themes (jobs, free, DIY) go to a shared list via
   create_negative_keyword_list. One-off terms go via
   add_negative_keywords at the ad group level.
2. Every write runs dry_run: true first. Show the diff, wait for my
   confirm, then apply.
3. Harvest additions are a separate confirm. Never bundle blocks and
   adds into one yes.

## Never
- Never negate brand or near-brand terms without asking.

4. Creative Fatigue Scan (Meta Ads)

Save as ~/.claude/skills/creative-fatigue-scan/SKILL.md

---
name: creative-fatigue-scan
description: Scan Meta campaigns for creative fatigue using frequency
  and CTR decline, then pause or rebalance through the PaidSync MCP
  after a dry_run preview and an explicit confirm.
---

# Creative Fatigue Scan (Meta Ads)

## Read
1. Run get_fb_campaign_performance for the last 14 days and the
   14 days before that.
2. For each campaign compare the two windows: frequency, CTR, CPA.
3. Flag as fatigued when frequency is above 4 and CTR fell 30
   percent or more, or CPA rose 40 percent or more on stable spend.

## Propose
A fatigue table: campaign, frequency, CTR change, CPA change,
verdict. Two verdicts only. Pause, for clear decay with spend
behind it. Rebalance, for campaigns worth keeping at lower budget
while new creative ships.

## Execute, gated
- Pause verdicts run pause_fb_campaign with dry_run: true first.
- Rebalance verdicts run update_fb_campaign with the new budget,
  dry_run: true first.
- Show every planned change, wait for my confirm, then apply.
- End the run with a plain list of which creative angles to brief
  next, based on what decayed slowest.

## Never
- Never pause a campaign still in its learning phase.
- Never touch campaigns spending under $10 a day. Noise, not
  signal.

5. LinkedIn Campaign QA

Save as ~/.claude/skills/linkedin-campaign-qa/SKILL.md

---
name: linkedin-campaign-qa
description: QA every live LinkedIn Ads campaign against a launch
  checklist, then fix failures through the PaidSync MCP after a
  dry_run preview and an explicit confirm.
---

# LinkedIn Campaign QA

## Read
1. Run get_campaign_performance for LinkedIn, last 30 days.
2. Pull each live campaign's settings through the LinkedIn tool
   group.

## Check every live campaign against this list
- Spend but zero leads or clicks for 14 days or more
- Audience so narrow it cannot exit learning, or so broad the
  targeting means nothing
- No end date on a campaign meant as a test
- Daily budget out of line with the campaign group's plan
- Duplicate campaigns competing for the same audience

## Propose
A QA table: campaign, checks failed, evidence, proposed fix. Fixes
are one of three moves. Adjust settings, adjust budget, or pause.

## Execute, gated
- Setting and budget fixes run update_linkedin_campaign,
  dry_run: true first.
- Dead campaigns run pause_linkedin_campaign, dry_run: true first.
- Show planned changes, wait for my confirm, apply, report.

## Never
- Never pause the only live campaign in a campaign group without
  flagging that the whole group goes dark.

6. Conversion Tracking Repair (GTM + GA4)

This is the one no report-only skill can touch. A document cannot re-publish a GTM container, and broken tracking silently corrupts every optimization decision the other five skills make. Run it first on any new account.

Save as ~/.claude/skills/conversion-tracking-repair/SKILL.md

---
name: conversion-tracking-repair
description: Find broken, missing, or duplicate conversion tracking
  across GTM and GA4, then repair tags and events and publish the
  container through the PaidSync MCP after explicit confirms.
---

# Conversion Tracking Repair (GTM + GA4)

The highest-stakes skill here. Broken tracking silently corrupts
every optimization decision the other skills make. Run this first
on any new account.

## Read
1. List every tag, trigger, and variable in the active GTM
   workspace.
2. List GA4 key events, data streams, and conversion settings.
3. Cross-reference both against the ad platforms' conversion
   actions.

## Hunt for these defects
- The same conversion firing from two tags. Doubled numbers.
- Tags with no trigger, or triggers pointing at retired pages.
- A GA4 key event the ad platform imports that no longer receives
  data.
- Purchase or lead events missing entirely on new landing pages.

## Propose
A repair plan. Each defect gets its evidence, its blast radius in
plain terms, and the exact fix.

## Execute, gated, two stages
1. Stage fixes in the GTM workspace: create_gtm_tag or
   update_gtm_tag for tag repairs, create_ga4_conversion_event for
   missing GA4 events. Each write runs dry_run: true first, then
   waits for my confirm.
2. Publishing is separate. Run create_gtm_version, then stop. Show
   me the version diff. publish_gtm_version runs only on a second
   explicit confirm, because publish is the moment changes go live
   on the site.

## Never
- Never publish a version containing changes I have not reviewed.
- Never delete tags. Pause by removing triggers. Deletion is a
  human decision.