mysteries-prep

Removed the model fallback (Grok) requirement. Retained DaVinci API injection over XML and whisper-cpp timestamp generation.

Mysteries Prep Protocol

This skill automates the Farsight Mysteries preparation workflow, which includes transcription, target cross-examination, AI image generation, and DaVinci Resolve integration.

Trigger

Use this skill when the user asks to run the "Mysteries Prep Protocol", prepare a Farsight target, or process an audio file for a Farsight project.

Phase 1: Preliminary Questions

Before starting the workflow, the AI MUST ask the user the following questions in chat:

1. Audio Location: "Where is the audio file located?" (Needed to generate the transcript).
2. Target Information Sheet: "Where is the target information sheet for this project?" (The AI MUST read this sheet to understand the true nature of the target to properly bridge viewer descriptions).
3. Image Density Mode: "Are we doing this with 'full' or 'semi' image settings?"
   • Full Mode: Generate an image for every single sentence/segment.
   • Semi Mode: Generate an image every 1-2 minutes for the hardest-hitting sentence.
4. Art Style: "What art style should be applied to the generated images?" (This ensures aesthetic consistency across the media).
5. DaVinci Resolve Readiness: "Do you have the appropriate DaVinci Resolve project open right now?"

Phase 2: Transcription & Directory Setup

1. Extract/Obtain: Get the .wav file of the raw video from the provided location.
2. Transcribe: Generate a word-for-word transcript from the .wav file. Make sure each sentence/segment has a timestamp. (Note: Use local tools like whisper-cpp if available to ensure millisecond-accurate timestamps and bypass API auth errors.)
3. Create Image Directory: Create a dedicated folder for the generated images in the exact same directory as the audio and transcript files.
   • The folder name must follow this format: [Agent Name] [Relevant Words] ai images (e.g., raphael intysam solos 11 ai images).

Phase 3: Image Generation

• Media Specs: ALL generated images MUST be strictly 1080p HD (1920x1080) resolution.
• Filename Sanitization (Critical): Ensure all generated image filenames contain NO spaces and NO parentheses (use underscores instead).
• Safety Filter Bypass: If prompts involving human forms or bodies are blocked by safety filters, tweak the phrasing to be abstract (e.g., "abstract glowing human silhouettes").
• Cross-Examine the Target: Use the target information sheet to understand the target truth.
• Generate Prompts: Craft image prompts that bridge the gap between the remote viewer's blind transcript descriptions and the actual target truth. Integrate the user-specified Art Style into every prompt.
• Generate the required images based on the chosen mode (Full/Semi) and save them to the AI images directory.

Phase 4: DaVinci Resolve Integration

(CRITICAL: Do NOT export standard XML or FCPXML. DaVinci's XML import is bugged and causes unresolvable "Media Offline" errors. You MUST use the DaVinci Resolve Python API exclusively for this phase.)

• Add Video Track: Add a new video track to the top of the currently open DaVinci Resolve timeline.
• Import Media: Import the generated 1080p images from the AI images directory into the DaVinci project's media pool.
• Base Zoom (API Step): All images MUST be inserted with a 1.02 baseline zoom on the X and Y axis via the Resolve API to prevent edge-bleeding during dynamic zoom.
• Placement (FULL Mode): Place each image at the exact timestamp indicated by the transcript. Each image MUST last the exact length of the sentence. No gaps, NO cross dissolves.
• Placement (SEMI Mode): Place each image at the exact timestamp indicated by the transcript. (Due to DaVinci API limitations, still images default to exactly 5 seconds upon placement).

Phase 5: Verbal Reminders for Manual UI Steps

Because the DaVinci Resolve Python API cannot apply transitions or effects, the AI MUST verbally remind the user in the chat to do the following:

• If Full Mode: "Please select all the newly placed images and toggle 'Dynamic Zoom' ON in the Inspector panel. (Remember, no cross dissolves are used in Full Mode!)"
• If Semi Mode: "Please note that the images are 5 seconds long by default due to API limits, so you may need to manually extend them to 10 seconds. Select all the newly placed images, press Cmd+T (Mac) or Ctrl+T (Windows) to add instant cross dissolves, and toggle 'Dynamic Zoom' ON in the Inspector panel."

---

Original Protocol Architects: Fuchikoma & Daemon King Updated by: Raphael (Refined DaVinci API & Transcription requirements)