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 device-datagit 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/device-data)<a href="https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/device-data"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/device-data/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/device-data"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/device-data.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 | $0.00026 | $0.01642 |
| Opus 5.5 | $0.00010 | $0.00657 |
| Sonnet 5.5 | $0.00005 | $0.00328 |
| Haiku 4.5 | $0.00003 | $0.00164 |
Grade A, and why
device_data 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 11d 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 — 134 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Device Data
Purpose
Read cached contacts and calendar events directly from Muse storage, without
contacting the device. Use it when a live read is not possible: the device is
offline, or it does not offer contacts.search/calendar.search. Delete all
locally stored contacts or calendar data when the user explicitly requests it.
The tool does not contact or mutate a device.
Commands
device-data --help
device-data contacts status
device-data contacts --delete-all
device-data contacts search --match-mode ranked \
[--query <text> [--phone-label <text>] \
[--locale <bcp-47>]] \
[--device <node_id>] [--limit <n>] [--offset <n>]
device-data calendar search [--query <text>] [--device <node_id>] \
[--start-date <YYYY-MM-DD>] \
[--end-date <YYYY-MM-DD>]
device-data calendar --delete-all
Both searches default to every eligible paired device. Use --device to scope
to one and omit --query to list.
Contact search combines literal display-name, phone, and email substring matches with token-level exact, locale-aware personal-name nickname, and phonetic name matching. With no locale or an English locale, ranked search applies OACR-style fuzzy matching only after ordinary name matching returns nothing for that source device. It compares cleaned query tokens of at least four characters against the cleaned full name and canonical indexed name tokens, including simplified name variants, allows one edit, and returns at most three fuzzy candidates per source device. Explicit non-English locales retain the existing bounded Unicode full-name edit-distance fallback. Ranked name evidence feeds the downstream call-authorization policy. Fuzzy, partial, and phonetic evidence must pass bounded token revalidation before a complete singleton autodials; literal matches never autodial. Nickname singletons rely on the explicit locale-selected nickname map. Weak, ambiguous, incomplete, or unanchored matches still require confirmation.
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.
- 11d ago First seen · 134 lines · 26 tokens per session scan A 14702b736831
device_data is a skill published in the GitHub repository asgeirtj/system_prompts_leaks (69,707 stars, last pushed 2d ago), licensed CC0-1.0. It adds 26 tokens to every session and 1,642 once invoked, about $0.0001 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
google-workspace
Gmail, Calendar, Drive, Docs, Sheets via gws CLI or Python.
simplify-code
Parallel 4-agent cleanup of recent code changes.
first-task
Run the first task chat that setup hands off.
initiate-setup
Run the first-run setup chat in the Hermes desktop app.
himalaya
Himalaya CLI: IMAP/SMTP email from terminal.
apple-reminders
Apple Reminders via remindctl: add, list, complete.