Skip to content

Prompt Decorator

The Prompt Decorator policy prepends or appends content to prompts before they are sent to the upstream LLM. Use it to inject system instructions, safety guidelines, or contextual information into every request — without requiring changes to the client application.

Configuration Parameters

Parameter Required Default Description
Decoration Yes — The content to inject. Either a plain text string (for single-field prompts) or a list of chat messages with role and content (for chat-format requests).
JSON Path Yes — A JSONPath expression pointing to the field in the request body where the decoration is applied (e.g., $.messages for a chat array, $.prompt for a text field).
Append No false When true, the decoration is appended after the existing content. When false (default), it is prepended before.

Decoration Formats

Text decoration — Use a plain string to inject text into a single-field prompt:

"decoration": "You are a helpful assistant. Always respond in formal English.\n\n"

Chat message decoration — Use an array of role/content objects to inject messages into a chat conversation:

"decoration": [
  {
    "role": "system",
    "content": "You are a helpful assistant. Only answer questions about software development."
  }
]

Add This Policy

  1. Navigate to AI Workspace > LLM Providers or LLM Proxies.
  2. Click on the provider or proxy name.
  3. Go to the Guardrails tab.
  4. Click + Add Guardrail and select Prompt Decorator from the sidebar.
  5. Enter the Decoration content.
  6. Set the JSON Path to the field in the request where content should be injected.
  7. Choose whether to Append (after) or prepend (before) the existing content.
  8. Click Add (for providers) or Submit (for proxies).
  9. Deploy the provider or proxy to apply the changes.

Example: Inject a System Prompt

The following configuration prepends a system instruction to every chat completion request:

Parameter Value
JSON Path $.messages
Decoration [{"role": "system", "content": "Always respond in plain text. Do not use markdown formatting."}]
Append false

Original request:

{
  "model": "gpt-4o",
  "messages": [
    {"role": "user", "content": "Explain recursion."}
  ]
}

Request after decoration (sent to LLM):

{
  "model": "gpt-4o",
  "messages": [
    {"role": "system", "content": "Always respond in plain text. Do not use markdown formatting."},
    {"role": "user", "content": "Explain recursion."}
  ]
}