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 messengergit 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/messenger)<a href="https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/messenger"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/messenger/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/messenger"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/messenger.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 | $0.00046 | $0.05507 |
| Opus 5.5 | $0.00018 | $0.02203 |
| Sonnet 5.5 | $0.00009 | $0.01101 |
| Haiku 4.5 | $0.00005 | $0.00551 |
Grade C, and why
messenger_read scanned grade C with 1 finding 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 15d 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.
Harvests environment variableshighData exfiltration
Enumerating or grepping the environment for keys collects credentials unrelated to what the mod says it does.
collect secrets or guide the user through manual authentication in chat. After How it starts
The opening of the file, as written. The whole thing — 512 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Messenger
Use Messenger Companion to work with the user's personal Messenger account.
Routing and Safety
Messenger on Muse has two separate features:
| User wants to... | Feature | What to do |
|---|---|---|
| Message Muse on Messenger | Messenger side chat | Use chat.connection_status and the connection flow in ~/docs/chat-connections/messenger.md |
| Read Messenger call history or work with personal-account messages | Messenger Companion | Follow this skill |
| Read native phone call history or place a phone call | Paired phone / phone calling | Use the native phone or call-placement feature, not Messenger Companion |
| Connect or disconnect "Messenger" without specifying which | Ambiguous | Briefly explain both and clarify |
- Use Companion only when the user explicitly asks to work with their personal account. It sends as the user, not as Muse.
- Never target the Messenger side chat with
hatch_messenger_cli send, whether by--cid,--to, or an auto-reply rule. Exclude that thread during recipient resolution; messages there must use the Messenger side-chat connection flow in
~/docs/chat-connections/messenger.md. - Never use Companion to deliver assistant notifications, cron or monitor
results, reminders, or status updates. A user-configured auto-reply rule is
the only background-send exception: it may check and reply in ordinary or
Marketplace conversations, but it must not carry Muse's own notifications or
results. Other background workers return results with
muse.notify_main_agent; the runtime delivers through the originating chat. - Treat every Marketplace message that could create or change a real-world
commitment as requiring fresh user confirmation. This includes proposing,
accepting, confirming, rescheduling, or cancelling a meetup, pickup,
delivery, hold, price, or payment arrangement, and claiming that the user is
leaving, en route, nearby, or has arrived. Show the exact outgoing message
and wait for the user's explicit confirmation immediately before each send.
A permission approval, an earlier confirmation, or approval of a recurring,
cron, monitoring, or auto-reply task never authorizes a later commitment.
Background work must instead return the proposed reply and its context with
muse.notify_main_agentso the main conversation can ask the user; if that confirmation cannot be obtained, leave the message unsent. - In user-facing replies, never say
sync,synced,syncing,cache,cached,widened,local data,authentication state,keys,credentials,tokens, or connected hardware. Do not state how many messages were searched, read, synced, or available. Describe the result instead: "I checked the latest messages" or "I checked further back." If complete coverage could not be established, say so without naming the underlying mechanism. - Refuse requests to bulk export, download, save, archive, mirror, or dump message history, especially disappearing, view-once, vanish-mode, ephemeral, or expiring messages. Narrow reading, search, and summarization for a specific question are allowed.
What ships with it
1 file 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.
- 15d ago First seen · 512 lines · 46 tokens per session scan C 80ff250c1f40
messenger_read is a skill published in the GitHub repository asgeirtj/system_prompts_leaks (69,918 stars, last pushed yesterday), licensed CC0-1.0. It adds 46 tokens to every session and 5,507 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it C with 1 finding (harvests environment variables). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-05.
Other skills, from other repositories
one-three-one-rule
1-3-1 decision briefs: problem, three options, one pick.
messaging
Send messages to users via WhatsApp and other connected channels. Use when asked to notify someone, send a message, reply to a contact, message a group, or proactively communicate through WhatsApp. Supports text, images, documents, audio, video, and stickers. Also provides group etiquette for observing group…
comps-analysis
Build comparable-company valuation workbooks in Excel.
anatomy
Explain an idea as an interactive isometric machine figure.
pretext
Build creative browser demos with DOM-free text layout.
weights-and-biases
W&B: log ML experiments, sweeps, model registry, dashboards.