SyssPrompt

Icon next to the "Share" button (DeepSeek/ChatGPT) or in the send bar (Claude) to mark which system prompt(s) to add to the first message of each new conversation

Aby zainstalować ten skrypt, wymagana jest instalacje jednego z następujących rozszerzeń: Tampermonkey, Greasemonkey lub Violentmonkey.

You will need to install an extension such as Tampermonkey or Violentmonkey to install this script.

Aby zainstalować ten skrypt, wymagana jest instalacje jednego z następujących rozszerzeń: Tampermonkey, Violentmonkey.

Aby zainstalować ten skrypt, wymagana będzie instalacja rozszerzenia Tampermonkey lub Userscripts.

You will need to install an extension such as Tampermonkey to install this script.

Aby zainstalować ten skrypt, musisz zainstalować rozszerzenie menedżera skryptów użytkownika.

(Mam już menedżera skryptów użytkownika, pozwól mi to zainstalować!)

You will need to install an extension such as Stylus to install this style.

You will need to install an extension such as Stylus to install this style.

You will need to install an extension such as Stylus to install this style.

Będziesz musiał zainstalować rozszerzenie menedżera stylów użytkownika, aby zainstalować ten styl.

Będziesz musiał zainstalować rozszerzenie menedżera stylów użytkownika, aby zainstalować ten styl.

Musisz zainstalować rozszerzenie menedżera stylów użytkownika, aby zainstalować ten styl.

(Mam już menedżera stylów użytkownika, pozwól mi to zainstalować!)

Autor
fjvico
Dziennych instalacji
0
Wszystkich instalacji
3
Oceny
0 0 0
Wersja
0.4
Utworzono
27-08-2026
Zaktualizowano
27-08-2026
Rozmiar
70,3 KB
Licencja
GPL-3.0
Dotyczy

SyssPrompt

Userscript (Tampermonkey / Violentmonkey) that adds a small 🐍 icon to DeepSeek, ChatGPT, and Claude from which you can mark which "system prompts" (behavior rules) you want to apply to the assistant, in the language of your choice. When you send your message, the script automatically adds the marked instructions just below, hidden beneath the scroll.

Current version: 0.3

Table of Contents

Installation

Easiest option (recommended for non-technical users)

  1. Open this page in your browser: SyssPrompt on Greasy Fork.
  2. Click the "Install this script" button. If this is your first time installing a userscript, the page itself will guide you to install the necessary extension first (Tampermonkey, Violentmonkey, etc., depending on your browser) — just follow the link it provides and try again.
  3. Confirm the installation in the dialog that opens from the extension. Done: no need to touch or copy any code.

Manual option (install from file)

  1. Install a userscript extension in your browser: Tampermonkey (Chrome, Firefox, Edge, Safari) or Violentmonkey.
  2. Create a new empty script from the extension's panel and paste the contents of deep_prompt.js, or open it directly if your manager allows it.
  3. Save. The script will activate automatically on:

No additional configuration or API keys are required.

Basic Usage

When you enter any of the three sites you will see a small 🐍 icon next to a two-letter code (the active language, e.g. ES):

  • DeepSeek: appears attached to the "Share" button in the top bar.
  • ChatGPT: appears fixed near the top-right corner.
  • Claude: appears fixed at the bottom right, next to the send bar.

Hovering over it (or clicking to keep it pinned open) displays a panel with the available rules. Check the ones you want to apply and type your message normally: just before sending (when pressing Enter or the send button), the script adds the checked instructions at the end of your message, after a line break, so they don't visually interrupt what you wrote.

  • DeepSeek/ChatGPT: the panel drops down.
  • Claude: the panel expands upward (the icon is at the very bottom).

How prompts are injected

The script remembers, for the current conversation, which rules were active the last time you sent a message. On each send it compares that state with the current one:

  • Newly checked rules: one sentence is added per rule, in this format:

From now on, follow the rule "Explain step by step": Explain your reasoning step by step, clearly and in detail, as if explaining it to someone without prior knowledge of the topic.

  • Newly unchecked rules: a single sentence grouping them all is added:

Cancel the rules "Explain step by step", "Be concise". The rest of the instructions I have given you remain in effect.

  • If nothing changed from the previous send (the same rules remain active), no new text is added — avoids repeating instructions already given in the same conversation.
  • If you mark no rules, the script does not touch your message.

Both templates are translated into the 13 supported languages and are applied in the language you have selected at the time of sending.

When a truly new conversation is detected (no previous messages), the record resets.

Languages

The icon shows the code of the active language. Clicking on those letters opens a dropdown with the 13 available languages, each written in itself:

Code Language
ES Español
CA Català
EU Euskara
GL Galego
EN English
FR Français
DE Deutsch
IT Italiano
PT Português
RU Русский
ZH 中文
JA 日本語
KA ქართული

When you choose a language from the dropdown:

  • The rule names and texts are displayed (and injected) in that language.
  • The checked state of rules is preserved (checked/unchecked state is independent of language: internal IDs are common across all languages).
  • The panel automatically returns to the rule list after choosing.
  • The default language is detected from the browser's language; if not among the supported ones, Spanish is used.

Each language includes, by default, four rules: Be concise, Explain step by step, Plain language, and Respond in [language] (to force the assistant to reply in that language even if you write to it in another).

Custom prompts and editing

Checking a rule on or off is done only by clicking its checkbox. Clicking the rule's name does something different: for your own custom rules, it loads that rule's name and text into the form at the bottom of the panel so you can edit it. The four built-in "factory" rules are read-only from the panel (their names are not clickable) — if you want a different version of one, create a personal copy as a custom rule and edit that instead of silently overwriting the original with no way to undo it.

  • Click a checkbox → the rule is added to / removed from your next message. Nothing else happens.
  • Click a custom rule's name → its name and text are copied into "Short name" and the instruction textarea, and the button changes to "✎ Update prompt". Edit the text and click that button to save the changes back onto that same rule (a "Cancel" link appears next to it to discard the edit instead).
  • Leave the form as-is (don't click any name first) and fill it in from scratch → clicking "+ Add prompt" creates a brand-new custom rule instead, checkable like any other, with an "×" to delete it.

Custom prompts are stored separately from the script's code, so they are not lost if you update the script to a new version in the future — only the "factory" prompts bundled in the file itself can change with an update.

Sharing the script

At the bottom of the panel there is a row of icons (inline SVG, wireframe style) to recommend SyssPrompt to others, all pointing to the Greasy Fork page:

  • 💬 WhatsApp, ✈️ Telegram, and ✉️ Email: open a direct link to each app with the message already drafted (translated into the active language).
  • 🔗 Copy link: copies the URL to the clipboard and shows a confirmation (✓).
  • Additional "more options" icon: only appears if the browser supports the native system sharing selector (e.g., Chrome/Edge on desktop or most mobile devices); Firefox on desktop does not support it, which is why the previous icons are always available as a universal alternative.

Site compatibility

Site Icon anchor Panel expansion
DeepSeek Next to the "Share" button (top bar) Downward
ChatGPT Fixed, top-right corner Downward
Claude Fixed bottom right (not inserted into the page's DOM) Upward

The script automatically detects the text editor of each site (<textarea> on DeepSeek, contenteditable ProseMirror-type editor on ChatGPT and Claude) and inserts the text appropriately for each case, including the special handling required for these editors to register the change correctly.

If the pages change their internal structure in the future, the icon may stop finding its "anchor button" and stay in its fixed fallback position (always visible) until the selectors are adjusted in the code.

Code customization

Inside the file you can adjust, among other things:

  • SITE_CONFIG: fixed position, expansion direction, and anchoring mode per site.
  • PROMPTS_BY_LANG: the "factory" rules for each language (same id across all languages).
  • UI_STRINGS: interface texts and the activation/deactivation templates, per language.
  • LANGUAGE_NATIVE_NAMES: the names displayed in the language dropdown.
  • SHARE_ICONS_SVG: the SVG icons in the "Share" row.
  • PROMPT_JOIN / SEPARATOR: how rules are separated from each other and from your message.

Storage and privacy

The script makes no network calls of its own and does not send data to any server (the "Share" links are the only ones that open an external URL, and only if you click on them). Everything it saves stays in your browser, via GM_setValue/GM_getValue (or localStorage if your manager does not support GM_*):

  • Which rules you have checked.
  • The selected language.
  • Your custom prompts.

This data is shared between DeepSeek, ChatGPT, and Claude because it is the same script installed, but it does not leave your machine.

Troubleshooting

  • The icon does not appear or stays in a fixed corner: the site's anchor button was not found; check the browser console (look for lines [SystemPrompt]) for more detail.
  • Prompts are not injected: check the console log input found: — if it points to a hidden or empty element, the text field selector may need adjustment (the site may have changed its HTML).
  • I want to see what's happening step by step: open developer tools (F12) → "Console" tab before sending a message; the script logs each step with the [SystemPrompt] prefix.

Support the project

If you find it useful, there is a small support link at the bottom of the panel (non-intrusive, makes no network calls and collects no data). If you prefer not to see it, you can safely delete the block marked SUPPORT ... end SUPPORT block in the code; the rest of the script continues to work exactly the same.

License

GPL-3.0. Author: Francisco Vico.