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 gemini-live-api-devgit 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/gemini-live-api-dev)<a href="https://agentmods.dev/skills/asgeirtj/system_prompts_leaks/gemini-live-api-dev"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/gemini-live-api-dev/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/gemini-live-api-dev"><img src="https://agentmods.dev/badge/skills/asgeirtj/system_prompts_leaks/gemini-live-api-dev.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 | $0.00110 | $0.04441 |
| Opus 5.5 | $0.00044 | $0.01776 |
| Sonnet 5.5 | $0.00022 | $0.00888 |
| Haiku 4.5 | $0.00011 | $0.00444 |
Grade A, and why
gemini-live-api-dev 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 today.
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.
Copies of this mod
1 near-identical copy found in the catalogue:
- gemini-live-api-dev — 92% identical, 20 lines differ
How it starts
The opening of the file, as written. The whole thing — 409 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Gemini Live API Development Skill
Overview
The Live API enables low-latency, real-time voice and video interactions with Gemini over WebSockets. It processes continuous streams of audio, video, or text to deliver immediate, human-like spoken responses and background reasoning.
Key capabilities:
- Bidirectional audio streaming — real-time mic-to-speaker conversations
- Background reasoning (extended thinking) — multi-step background reasoning with spoken conversational fillers
- Live streaming transcription — real-time speech-to-text with interim and finalized streams
- Video streaming — send camera/screen frames alongside audio
- Text input/output — send and receive text within a live session
- Audio transcriptions — get text transcripts of both input and output audio
- Voice Activity Detection (VAD) — automatic server VAD, client-side Hybrid VAD, and manual Push-to-Talk
- Asynchronous function calling — non-blocking tool execution while audio continues streaming
- Full-session client content — inject and update conversation turns mid-stream
- Session management — context compression, session resumption, GoAway signals
- Ephemeral tokens — secure client-side authentication
[!NOTE]
The Live API connects directly via WebSockets. For WebRTC support or simplified integration, use a partner integration.
Models
Current Models (Use These)
gemini-3.8-live— Default option for most low-latency voice agent experiences and real-time dialogue without reasoning delays. Supports interleaved reasoning, asynchronous function calling by default (behavior: NON_BLOCKING), and full-session client content updates.gemini-3.8-live-extended-thinking— High-reasoning audio-to-audio model recommended when higher background reasoning is required during live interactions. Processes background reasoning and async tool calls (behavior: NON_BLOCKINGrequired) while streaming continuous spoken conversational fillers; lifecycle managed viainteraction_status(IN_PROGRESSvsIDLE).gemini-3.5-transcribe-live— Real-time streaming speech-to-text with interim hypotheses, finalized transcripts, smart formatting, and Hybrid VAD.gemini-3.5-live-translate-preview— Real-time speech-to-speech streaming translation across 70+ languages.
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.
- today First seen · 409 lines · 110 tokens per session scan A e429015842dd
gemini-live-api-dev is a skill published in the GitHub repository asgeirtj/system_prompts_leaks (70,425 stars, last pushed today), licensed CC0-1.0. It adds 110 tokens to every session and 4,441 once invoked, about $0.0004 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 2026-10-30.
Other skills, from other repositories
serving-llms-vllm
Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.
saelens
Train sparse autoencoders to interpret model features.
telephony
Provision Twilio numbers, SMS/MMS, and AI outbound calls.
accelerate
Run PyTorch training across GPUs with minimal changes.
flash-attention
Speed up long-sequence transformer training and inference.
nemo-curator
Curate LLM training data: dedupe, filter, PII redaction.