Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add swan-gtm/gtm-skills --skill google-ads-write-safetygit clone --depth 1 https://github.com/swan-gtm/gtm-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/swan-gtm/gtm-skills/google-ads-write-safety)<a href="https://agentmods.dev/skills/swan-gtm/gtm-skills/google-ads-write-safety"><img src="https://agentmods.dev/badge/skills/swan-gtm/gtm-skills/google-ads-write-safety/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/swan-gtm/gtm-skills/google-ads-write-safety"><img src="https://agentmods.dev/badge/skills/swan-gtm/gtm-skills/google-ads-write-safety.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00107 | $0.00871 |
| Opus 5 | $0.00053 | $0.00436 |
| Sonnet 5 | $0.00021 | $0.00174 |
| Haiku 4.5 | $0.00011 | $0.00087 |
Grade A, and why
google-ads-write-safety scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 71 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Write Safety for AI Agents on Google Ads Accounts
Applies the moment an agent moves from reading an account to writing to it. The account is real money and the advertiser's livelihood — a hallucinated success message isn't a rough edge, it's a trust catastrophe the first time it happens.
The play
- Classify the request by blast radius before touching anything: read-only, low-risk (a single keyword, negative, or bid move within a tight band), or account-level (budget swings, conversion tracking, campaign creation or deletion).
- Default to propose-only — pull data, calculate, recommend, but execute nothing without an explicit go-ahead. An authorization covers the specific action just approved, never the rest of the session.
- Keep the low-risk band actually low-risk: bid moves within roughly ±20%, negatives only against clear waste (spend with zero conversions), no whole-campaign pauses, no budget swings past ~20%, and never touch conversion actions or billing without a separate, explicit approval.
- Before executing, show a preview — current value, proposed value, and what triggered it (the search term, the CPA, the pacing). New campaigns and new ads get created paused, full stop, so a human reviews before spend starts.
- Require an explicit confirmation tied to that specific preview. "Go ahead," "do it," "approved" counts; a related-but-different message, or silence, does not. Batch operations still get one itemized preview — never a blanket yes for a list the human hasn't actually read.
- Execute, then produce a receipt: the operation, the exact resource changed, the before/after state, and how to reverse it. If nothing else survives, the receipt should.
- If the write failed, partially applied, or the tool was never called, say so in plain language immediately. Never generate a receipt for something that didn't happen, and never imply full success when only part of a batch went through.
What good looks like
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 71 lines · 107 tokens per session scan A 2a6d0cde2c9a
google-ads-write-safety is a skill published in the GitHub repository swan-gtm/gtm-skills (153 stars, last pushed 2d ago), licensed MIT. It adds 107 tokens to every session and 871 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
sector-rotation
An analysis framework for comparing industries in the Chinese A-share stock market, using business conditions, price momentum, valuation, and money flows. It produces rankings and higher- or lower-allocation suggestions.
strategy-pivot-designer
Detect backtest iteration stagnation and generate structurally different strategy pivot proposals when parameter tuning reaches a local optimum.
twitter-reader
Read Twitter/X for financial research using opencli (read-only). Use this skill whenever the user wants to read their Twitter feed, search for financial tweets, view bookmarks, look up user profiles, or gather market sentiment from Twitter/X. Triggers include: "check my feed", "search Twitter for", "show my…
chenhao-limit-up
A framework for judging Chinese A-share stocks that have reached the daily price-rise limit, using market mood, sector leadership, and trading momentum.
furusato
A Japanese hometown-tax donation manager for furusato nozei, a system where donations to municipalities can qualify for an income-tax or local-tax deduction. It reads donation receipts, stores donation records, and calculates deduction limits.
reading-receipt
An image-reading workflow for extracting structured information from receipts, invoices, and hometown-tax donation certificates. It can first extract text from PDFs and otherwise read their images.