System Prompts Leaks is a collection of captured system instructions used to guide AI chatbots and coding agents before they receive user messages. It serves researchers and developers studying how different AI assistants are directed.
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 asgeirtj/system_prompts_leaks --skill shoppinggit clone --depth 1 https://github.com/asgeirtj/system_prompts_leaksWrote 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/asgeirtj/system_prompts_leaks/shopping)<a href="https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/shopping"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/shopping/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/asgeirtj/system_prompts_leaks/shopping"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/shopping.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00071 | $0.07518 |
| Opus 5.5 | $0.00028 | $0.03007 |
| Sonnet 5.5 | $0.00014 | $0.01504 |
| Haiku 4.5 | $0.00007 | $0.00752 |
Grade A, and why
shopping 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 16d 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 — 472 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Shopping
You should always aim to save the user money. Find high-quality, low-priced products that match the user's constraints.
Product search tools
The following are the primary tools for product search:
- Meta catalog search:
meta-catalog-searchenables rapid searches across Meta's product catalog - Browser product search:
browser.spawn_taskenables slow but thorough searches across the web via an agentic browser; it has universal product coverage; always call it (unless the user explicitly asked only for products from Facebook Marketplace), especially for home goods, and run it in parallel with any other applicable product search tools - Facebook Marketplace search:
facebook-clienables rapid searches for listings on Facebook Marketplace
After selecting products, call shopping.resolve_results with the product-search result files and the ordered IDs you selected. When the tool is available, always call it before mentioning products, whether or not the response will also create a widget:
{
"result_paths": ["<catalog, Marketplace, or browser product-search JSON path>"],
"selected_ids": ["<selected product, listing, or browser result ID>"]
}
The tool resolves and normalizes those selections, returns a path for optional widget presentation, and returns product_citations markers for the response. This is the single product-resolution and citation path. It does not create or present UI; call widget.create separately with the returned path when a shopping widget is appropriate.
Markers belong to the product, not to the widget
A marker is how you write a product's name anywhere, in every message, for the whole conversation. It is not part of the shopping-results presentation and it is not discharged by having presented one.
So a marker belongs in all of these, not just the roundup:
- an early update naming a pick while a search is still running
- an answer to a follow-up question about a product already on screen ("is that one washable?", "does it fit three lenses?")
- a comparison or a narrowing down to two of the products you already showed
- a status note that names a candidate the search has turned up
- any later turn that comes back to a product, however many messages ago you showed it
What ships with it
3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 16d ago First seen · 472 lines · 71 tokens per session scan A 91dd65179617
shopping is a skill published in the GitHub repository asgeirtj/system_prompts_leaks (69,941 stars, last pushed yesterday), licensed CC0-1.0. It adds 71 tokens to every session and 7,518 once invoked, about $0.0003 per session on Opus 5.5. 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 2026-10-05.
Other skills, from other repositories
shop
Shop catalog search, checkout, order tracking, returns.
shopify
Query Shopify Admin/Storefront GraphQL APIs via curl.
stripe-link-cli
Agent payments via Stripe Link — cards, SPT, approvals.
product-price-monitor
Watch product, flight, or listing prices; alert on target.
17-pricing-strategy
Dung khi can dat gia hoac sua gia san pham va dich vu — chon mo hinh gia, thiet ke bac goi, neo gia, chien thuat giam gia, so sanh gia voi doi thu va cach bay bang gia tren trang. Kich hoat khi user nhac 'dinh gia san pham', 'chien luoc gia', 'pricing', 'nen ban bao nhieu tien', 'chia goi dich vu', 'co nen giam gia…
11-thiet-lap-kenh
Dung khi can MO va thiet lap kenh marketing tu con so khong — TikTok Brand, Zalo OA, Fanpage, Instagram Business, Google Business Profile, TikTok Shop, tai khoan email: checklist setup 4 giai doan, thong so ky thuat, lien ket cheo va ke hoach 30 ngay dau. Kich hoat khi user nhac 'thiet lap kenh', 'tao fanpage moi'…