Every published Claude ads skill we checked live on August 1, 2026 stops at a report. The six skills below finish the job. Each one reads live account data, proposes a specific change, then executes it through the PaidSync MCP behind a dry_run preview and your explicit confirm.
A Claude skill is a SKILL.md file: YAML frontmatter between two --- markers with a name and description, then markdown instructions Claude follows when the skill runs. Install each one in its own folder under ~/.claude/skills and Claude loads it when relevant, or when you type the slash command. The format follows the Agent Skills open standard. All six below are packaged as installable skills, each with its own page on paidsync.ai/skills.
Install All Six Skills in One Command
The skills live at paidsync.ai, so the open source skills CLI can install them straight from the site. Run this from any project folder:
npx skills add https://paidsync.aiThe CLI reads the skill index at paidsync.ai/.well-known/skills/index.json, checks every file against its sha256 digest, and asks which agents to install into. Add -a claude-code to install straight into Claude Code, or --skill wasted-spend-hunter to install one skill. Every skill runs on the PaidSync MCP server at https://mcp.paidsync.ai/mcp, so connect it first. Prefer to read before you install? Each skill below links to its full SKILL.md.
Where Every Published Ads Skill Stops
We checked the most-cited public Claude ads skill libraries live on August 1, 2026. They are well built. They also all end at the same place, which is a document.
Ryze AI's 15 Claude skills for Google Ads end in outputs like a ranked diagnosis or, for the negative keyword skill, "an import-ready list formatted for Google Ads Editor." You do the importing.
Checked live August 1, 2026
Their 10-skill download set delivers "ready-to-paste" lists and instructs Claude to "never auto-execute budget changes." Recommendations, then a human applies them.
Checked live August 1, 2026
Their 30-skill library, updated July 19, 2026, says it plainly: every skill "is written to run read-only," and outputs are "recommendations, exclusion lists, and drafts you apply yourself."
Checked live August 1, 2026
That is the correct design for a read-only connection. Google's own open source Google Ads API MCP server, announced October 7, 2025, exposes read tools only, three of them, as of August 1, 2026. When the pipe only reads, every skill built on it must end at a report. The report still leaves you with the worst part of the job: logging into three ad platforms and typing the changes in yourself.
PaidSync's MCP writes. 630+ tools across 14 platforms, with full write on Google Ads, Meta and LinkedIn, the full Tag Manager tag, trigger and variable lifecycle including container publish, and GA4 configuration. With a write-capable connection, a skill can be written to finish what it finds.
How the Execution Gate Works
Skills that write need a leash, so every skill below ends with the same three steps and never skips them.
- Preview. The write tool runs with
dry_run: truefirst wherever the tool supports it, returning the planned change without applying it. A tool with no dry run is shown to you with its exact arguments before it is called. - Confirm. Claude shows you the exact planned change and waits for an explicit yes. Silence is not a yes. Over MCP, PaidSync writes no approval card, so the skill makes Claude the approval step, and destructive operations need
confirm_destructive. - Apply and report. Only after your yes does the real call run. The skill closes by reporting what changed. PaidSync logs applied changes, on a best-effort basis, with a 7-day before-and-after outcome.
You can revoke a PaidSync session at any time, which cuts off write access mid-conversation if you ever want the hard stop. This gate is not a nice-to-have. It is why six agencies became paying PaidSync customers in the first four days after launch in May 2026. Nobody hands an AI write access to client budgets without a preview and a confirm in the loop.
The Six Skills
Each skill lists what it reads, what change it proposes and how it applies it, with its install command and a link to its full page and SKILL.md. Thresholds like a $25 spend floor are starting defaults, not benchmarks. Each skill asks once per account before it uses them.
1. Wasted Spend Hunter (Google Ads)
get_wasted_spend_report and get_search_terms_report for the last 30 days, a 90-day search terms check, and the negatives already in place.add_negative_keywords, first with dry_run: true, then live only after your yes. At most 50 negatives in one approved batch.Install with npx skills add https://paidsync.ai --skill wasted-spend-hunter
Read the full Wasted Spend Hunter skill, or download its SKILL.md.
2. Negative Keyword Triage (Google Ads)
get_search_terms_report for the last 7 days, get_keyword_performance for 30 days of ad group context, and list_negative_keywords with list_shared_sets for what is already blocked.add_negative_keywords, the shared list tools (create_negative_keyword_list, add_to_negative_keyword_list, apply_negative_keyword_list) and add_keywords with match type EXACT. Blocks and harvests are two separate approvals.Install with npx skills add https://paidsync.ai --skill negative-keyword-triage
Read the full Negative Keyword Triage skill, or download its SKILL.md.
3. Meta Ads Audit
get_fb_account_info, list_fb_pixels, get_fb_pixel_events, list_fb_custom_conversions, list_fb_campaigns, list_fb_adsets, the ad set and ad performance reports, the placement, device, age, gender and country breakdowns, and get_fb_account_activity.pause_fb_ad, pause_fb_adset or update_fb_adset, previewed as a dry run first.Install with npx skills add https://paidsync.ai --skill meta-ads-audit
Read the full Meta Ads Audit skill, or download its SKILL.md.
4. Creative Fatigue Scan (Meta Ads)
get_fb_ad_performance for two adjacent 14-day windows, and list_fb_ads for each ad's name, ad set and creative.pause_fb_ad for pauses and update_fb_adset with a new daily budget for rebalances, each previewed as a dry run and applied after your yes.Install with npx skills add https://paidsync.ai --skill creative-fatigue-scan
Read the full Creative Fatigue Scan skill, or download its SKILL.md.
5. LinkedIn Campaign QA
list_linkedin_campaigns and list_linkedin_campaign_groups, get_linkedin_performance for 14 and 30 days, get_linkedin_campaign_targeting with get_linkedin_audience_size for each live campaign, and list_linkedin_conversions.update_linkedin_campaign for settings and budgets and pause_linkedin_campaign for dead campaigns, each previewed as a dry run first.Install with npx skills add https://paidsync.ai --skill linkedin-campaign-qa
Read the full LinkedIn Campaign QA skill, or download its SKILL.md.
6. Tracking Health (Tag Manager, GA4, Google Ads, Meta)
create_gtm_tag with a dry run, update_gtm_tag and create_ga4_conversion_event after an explicit yes, then create_gtm_version. publish_gtm_version runs only on a second yes.Install with npx skills add https://paidsync.ai --skill tracking-health
Read the full Tracking Health skill, or download its SKILL.md.
Connect an account and run the first skill in minutes. Free tier includes 15 tasks a month, no card.
Get started for free Browse the skill pagesWhat These Skills Do Not Do
The honest edges, as of September 23, 2026
- No campaign building. These six find problems and fix what exists. None of them creates campaigns, on any platform.
- No budget pacing. An earlier version of this page carried a Budget Pacing Checker. It is not part of the installable set.
- Search Console is read-only. 6 tools, all reads. Nothing here writes to GSC because nothing can.
- No unattended runs. Every skill waits for your confirm before any write. If you want ads work that runs without you in the loop, that is a different product decision and this page is not selling it.
- Thresholds are defaults, not benchmarks. The $25 spend floor, the frequency-above-4 flag, the 50,000-member audience floor. Starting points to tune per account, not industry research.
What Running These Costs
A single skill run makes a handful of tasks. A read or two, a dry_run, an apply. PaidSync's Free tier is $0 with 15 tasks a month, enough to connect an account and test a skill end to end. Pro starts at $99 a month for 600 tasks, with 1,200 tasks at $149 and 4,000 at $199. Team starts at $249 a month, and Done For You is custom. Full details on the pricing page.
Download the Skill Files
Each skill is one SKILL.md file, and each has its own page with the full file, a download link and its install command. The skills hub lists all six. For installers and agents, the index at /.well-known/skills/index.json follows the agentskills.io discovery schema and lists every file with a sha256 digest, which the skills CLI checks before it installs anything.
- Wasted Spend Hunter for Google Ads
- Negative Keyword Triage for Google Ads
- Meta Ads Audit
- Creative Fatigue Scan for Meta Ads
- LinkedIn Campaign QA
- Tracking Health for Tag Manager, GA4, Google Ads and the Meta pixel
Frequently Asked Questions
What is a Claude skill for ads?
A Claude skill is a SKILL.md markdown file with YAML frontmatter (a name and a description between two --- markers) that teaches Claude a repeatable procedure. Install all six from this page with npx skills add https://paidsync.ai, or save each one in its own folder under ~/.claude/skills. Claude loads a skill when it is relevant, or when you type its slash command. Skills follow the Agent Skills open standard. For ads, a skill defines what data to read, what change to propose, and how to apply it. The six on this page run on the PaidSync MCP server, so the last step is an applied change, not a to-do list.
Can these skills change my ad account without asking?
Not by design. Every skill on this page previews a write as a dry run where the tool supports one, shows the planned change, and waits for an explicit yes. Silence is not a yes. Over MCP, PaidSync writes no approval card, so the skill makes your assistant the approval step, and destructive operations need confirm_destructive. You can also revoke a session at any time.
Which platforms can these skills write to?
The six skills write to Google Ads, Meta and LinkedIn, to Google Tag Manager, including a container publish on a second yes, and to GA4 key events. PaidSync itself has 630+ tools across 14 platforms, with campaign creation on Google Ads, Meta, LinkedIn, ChatGPT Ads and Microsoft Ads.
Do the skills work on TikTok?
None of the six targets TikTok yet. They cover Google Ads, Meta, LinkedIn, Tag Manager and GA4. TikTok connects through the same PaidSync MCP server, so your agent can still pause ads, change budgets and pull TikTok reports there without a skill.
How are these different from other published Claude ads skills?
The published libraries we checked live on August 1, 2026 all end at analysis. Ryze AI's Google Ads skills end with output like "an import-ready list formatted for Google Ads Editor," their 10-skill set instructs Claude to "never auto-execute budget changes," and their 30-skill library states every skill "is written to run read-only." That is the correct design for read-only connections. The six skills here are written for a write-capable MCP, so each one ends in an applied change behind a dry_run preview and an explicit confirm.
Do I need PaidSync to use these skills?
Yes. Each skill names PaidSync MCP tools and tells the agent to stop if they are missing, because a skill can only apply changes when its connection can write. PaidSync's Free tier is $0 with 15 tasks a month, enough to connect an account and test a skill run. Pro starts at $99 a month for 600 tasks, with 1,200 tasks at $149 and 4,000 at $199. Team starts at $249 a month. Details on the pricing page.