AI - Content Prompt sub-action: default example prompt references nonexistent ${fieldContent} and fails

[mboutillette-dotcms]

Problem

The default example prompt shipped with the AI - Content Prompt workflow sub-action references ${fieldContent}, which is not a variable that exists in the Velocity context. Running the sub-action with the out-of-the-box prompt fails.

Current default (OpenAIContentPromptActionlet.java#L42):

We need an attractive search result in Google. Return a json object that includes the fields "pageTitle" for a meta title of less than 55 characters and "metaDescription" for the meta description of less than 300 characters using this content:\n\n${fieldContent}\n\n

Root cause

OpenAIContentPromptRunner.openAIRequest() evaluates the prompt with VelocityUtil.eval() against the context built by VelocityContextFactory.getMockContext(). That context provides $contentletToString, $contentMap, $dotContentMap and $contentlet — there is no fieldContent. Velocity renders the unresolved reference verbatim, so the model receives the literal string ${fieldContent} and no actual content. It replies with prose rather than JSON.

parseJsonResponse() then does:

response.substring(response.indexOf("{"), response.lastIndexOf("}") + 1)

With no { in the reply, indexOf returns -1 and lastIndexOf("}") + 1 returns 0, producing:

Unable to execute the Workflow Action. Error: Range [-1, 0) out of bounds for length 341

The default is also inconsistent with the sub-action's own getHowTo(), which correctly documents $dotContentMap. The literal \n\n in the default is a second (cosmetic) problem — it renders as backslash-n in the textarea rather than as line breaks.

Suggested fix

Update the default value to:

Return ONLY a raw JSON object — no markdown code fences, no commentary — with exactly two keys: "pageTitle" (a meta title under 55 characters) and "metaDescription" (a meta description under 300 characters), written to be an attractive Google search result for this content:

$contentletToString

Secondary

parseJsonResponse() should guard against indexOf("{") == -1 and throw BadAIJsonFormatException with a message naming the real problem (the AI did not return JSON, plus the response), instead of surfacing an opaque Range [-1, 0) error to the user.