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 forgetgit 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/forget)<a href="https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/forget"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/forget/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/forget"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/forget.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 | $0.00074 | $0.01374 |
| Opus 5.5 | $0.00030 | $0.00550 |
| Sonnet 5.5 | $0.00015 | $0.00275 |
| Haiku 4.5 | $0.00007 | $0.00137 |
Grade A, and why
forget 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 3d 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 — 122 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Forget
Treat forgetting as cleanup across Muse, not as editing one memory note. Information may also live in conversation history, other notes, preferences, goals, scheduled work, created items, search results, or active work that can write it back.
Use everyday language with the user and do not expose internal machinery. Say "memory," "reminders," "created items," "shared items," "logs," or "backups" instead of file names, database tables, indexes, projections, telemetry systems, tool names, agent types, or runtime machinery. Keep exact locators and technical details inside the private work. Get technical only when the user does or when they explicitly ask how the cleanup works.
Make a plan
Call forget.plan once with {}. It starts an ordinary untyped subagent with
the normal tool catalog that derives the subject from the current conversation,
so do not repeat sensitive text in tool arguments. Wait for its handoff instead
of polling or doing a parallel search. Muse pins this child to muse-special, or
to private Avocado on a confidential VM, independently of the active model
selection.
The planner reads references/artifact-inventory.md, stays read-only, and checks:
- original places and copies with the same meaning;
- memory, search results, conversation context, and summaries made from it;
- reminders, scheduled or active work, and outside sources that can recreate the information;
- shared or published copies and anything Muse cannot erase.
It returns a concise plan covering what it found, what should change, how it will verify the cleanup, and known limits. Keep the handoff discreet; refer to "that information" instead of copying it into a new memory, todo, filename, or report. Use conversations privately to find downstream copies and future activity, but do not present the continued visibility or retention of the chat itself as a cleanup limit.
Ask, then execute
Show the user the plan's scope, irreversible actions, and important limits, translated into the everyday categories above, then ask one direct confirmation question. Silence, ambiguity, partial approval, or a changed scope is not confirmation.
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.
- 3d ago First seen · 122 lines · 74 tokens per session scan A 3ffe3787b0d1
forget is a skill published in the GitHub repository asgeirtj/system_prompts_leaks (69,084 stars, last pushed today), licensed CC0-1.0. It adds 74 tokens to every session and 1,374 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
honcho
Configure and troubleshoot Honcho memory for Hermes.
pinecone-research
Agent RAG and long-term memory with Pinecone.
llm-wiki
Karpathy's LLM Wiki: build/query interlinked markdown KB.
qmd
Hybrid local search over notes, docs, and transcripts.
ai-persistence
Durability and state persistence for TanStack AI chats with @tanstack/ai-persistence. Routes to server chat persistence (withPersistence), client persistence (localStorage/IndexedDB), the store contracts, and adapter recipes. Distinguishes delivery durability (resumable streams) from conversation state. Use when…
ai-persistence/server
Server chat state with withPersistence from @tanstack/ai-persistence. Authoritative transcript, run lifecycle, durable interrupts/approvals, chatParamsFromRequest, reconstructChat, snapshotStreaming. Use when the server owns history, multi-device, or durable tool approvals. NOT client localStorage (see…