AiBatCustomPrmpt

This repository contains custom prompts for AI Bat.

The GH pages url is https://adobe-rnd.github.io/aem-assets-autometadata-tool/

Getting Started

1. Clone this repository
2. Enter the config (get from a peer on the project)
3. Add your custom properties and prompts
4. Test and iterate

Confidence Score, Reason, and Data Shape

The UI shows a confidence pill next to each metadata property. Values, scores, and reasons are stored in the following shape:

{ "<property>": { "value": "...", "confidence_score": 0.0..1.0, "confidence_reason": "..." } }

Fields:

• value: The generated metadata value.
• confidence_score: A number between 0 and 1 indicating how confident the AI is in the accuracy of the value.
• confidence_reason: A brief explanation (1-2 sentences) of why the confidence score is what it is.

Parsing behavior:

• If the model returns a JSON object shaped like { <property>: <value>, confidence_score: <0..1>, confidence_reason: "..." }, that is used directly.
• If the model returns legacy JSON (e.g., Title/Description/Keywords), values are wrapped as { value, confidence_score: null, confidence_reason: null }.
• If the model returns plain text (not JSON), the text is wrapped as { <property>: { value: text, confidence_score: null, confidence_reason: null } } for per-property calls (or mapped to defaults).

Exports:

• JSON export normalizes all properties to { value, confidence_score, confidence_reason }.
• CSV export includes three columns per property: value, confidence (raw score 0..1), and reason.

UI:

• Confidence pills display percentage (e.g., 82%) and use color tiers for low/medium/high.
• Hovering over a confidence pill shows the confidence reason as a tooltip.

System Prompt

You can now set a global System Prompt that is included in every Azure OpenAI request as a role: system message.

How to use:

1. Open the app and click 🧩 System Prompt (next to ⚙️ Configure LLM API).
2. Enter your system-level guidance (e.g., tone, safety, formatting rules) and click Save.
3. The value is stored in localStorage and automatically applied to all requests.

Notes:

• A default system prompt is included that enforces JSON output with confidence scores.
• Clicking "Clear" resets to the default system prompt (does not remove it).
• To truly disable the system prompt, manually clear the text and click Save.
• The existing Brand Prompt continues to prepend to the user prompt text. Both can be used together.

Contributing

Please follow the contribution guidelines when adding new prompts.