JS-Slash-Runner (Tavern-Helper)

Tavern-Helper is a SillyTavern extension that allows you to run external JavaScript code. Because SillyTavern does not support direct execution of arbitrary JavaScript by default, this extension uses an iframe to isolate and execute scripts, enabling them to run in certain restricted contexts.

WARNING

Executing custom JavaScript code carries security risks. Malicious scripts could steal API keys, chat logs, or sensitive information, or modify/damage your SillyTavern settings. Always inspect script content and ensure the source is trusted before execution.

The Prompt Viewer is designed to always show the most recent prompts by intercepting generation requests. It updates its view in three scenarios:

1. Opening the Viewer: Upon opening, the viewer triggers a fake generation request (simulating a click on the send button) to intercept and display the current prompts without actually sending them to the AI.
2. Manual Refresh: Clicking the refresh button (<i class="fa-solid fa-rotate-right"></i>) triggers a fake generation request to fetch the latest state.
3. Actual Generation: The viewer listens to all generation requests. It will automatically update whenever a prompt is sent via:
   • Clicking the send button (<i class="fa-solid fa-paper-plane"></i>).
   • Using the generate or generateRaw functions from Tavern Helper.
   • Any other method used to send prompts through SillyTavern.

To ensure compatibility with modern character cards, use the following recommended global settings for Worldbooks. Most current character card authors use these settings by default, so manual modification is generally unnecessary unless you are working with older content.

Recommended Settings:

• Scan Depth (扫描深度): 2
• Context Percentage (上下文百分比): 100
• Token Budget Cap (Token 预算上限): 0
• Minimum Activations (最小激活数): 0
• Max Depth (最大深度): 0
• Max Recursive Depth (最大递归深度): 0
• Insertion Strategy (插入策略): Character Worldbook Priority (角色世界书优先)
• Include Names (包括名称): false
• Recursive Scan (递归扫描): true
• Case Sensitive (区分大小写): false
• Match Whole Word (匹配整个单词): false
• Use Group Scoring (使用群组评分): false
• Overflow Alert (溢出警报): false

The # Msg. to Load setting (now referred to as # Msg. to Render) in the top bar controls the maximum number of messages displayed during gameplay. This optimizes performance by ensuring the tavern only renders a specific number of messages at any given time.

Behavior

• Limit Enforcement: If set to 5, the page will display at most 5 messages.
• New Messages/Replies: When a new message or reply is sent, the oldest message is removed from the render queue to make room for the new one (e.g., messages 4,5,6,7,8 become 5,6,7,8,9).
• Message Deletion: When a message is deleted, the system automatically fills the gap by rendering the next available old message (e.g., messages 5,6,7,8,9 become 4,5,6,7,8).

Configuration

• You can set this to any non-negative integer.
• For example, setting it to 1 will only display the single most recent message.

To build this project from source, you must have Node.js 22+ and pnpm installed.

Follow these steps to set up the environment and compile the project:

1. Install pnpm globally:

   npm install -g pnpm

2. Install project dependencies:

   pnpm install

3. Build the project:

   pnpm build

Alternatively, you can use pnpm watch to continuously monitor code changes. After making changes, simply refresh the SillyTavern web page to use the latest plugin code.

Enable the Maximize preset context length feature to lock the Context Size (tokens) setting in the preset panel to its maximum value (200w).

This prevents SillyTavern from incorrectly truncating prompts. Because SillyTavern's token calculation is often inaccurate—especially when extensions like Tavern helper macros or ST-Prompt-Template are processing the prompt—the calculated token count can be much higher than the actual count. Locking the context size to the maximum ensures that the full prompt is sent to the AI without being prematurely cut off by SillyTavern's internal limits.

By default, the standard Tavern 'Replace/Update Character Card' feature only updates character-specific metadata (such as avatar, description, and the first message) but fails to update previously imported World Info (Worldbook) data.

To ensure that World Info is also synchronized when updating a character card, enable the better_character_update optimization. When this option is active, the Replace/Update Character Card function will simultaneously update both the character card information and its associated World Info.

If you are using older Worldbooks that are incompatible with the recommended settings, you may need to disable the recursive scan feature.

To do this, use the built-in script named "Disable recursion of worldbook entries in one click" located in the project's built-in library to process the entries.

Enable this feature to lock the Context Length (in tokens) in the preset panel to its maximum value (2,000,000).

By default, Tavern applies a limit to the number of tokens sent to the AI. If the prompt exceeds this limit, Tavern truncates it. Because Tavern's token calculation is often imprecise and can be inflated by plugin processing (such as Tavern-Helper macro replacements or conditional prompt templates), the calculated token count is frequently much higher than the actual count.

Locking the context length to the maximum prevents Tavern from incorrectly truncating prompts that could otherwise be sent in full to the AI, and it does not negatively impact gameplay.