prompt-library-audit-and-rebuild

name	prompt-library-audit-and-rebuild
description	Automated workflow for auditing and mass-refining large-scale AI prompt libraries (JSON + Markdown) to ensure semantic character consistency and technical compliance.

prompt-library-audit-and-rebuild

Automated workflow for auditing and mass-refining large-scale AI prompt libraries (JSON + Markdown) to ensure semantic character consistency and technical compliance.

Trigger Conditions

• After a structural rebuild of a prompt library where content integrity is suspect.
• When "generic" terms (e.g., 'a woman', 'a dog') are detected in production prompts.
• Before initiating high-volume ComfyUI batch runs to ensure character/gear anchors are present.

Workflow Steps

1. Audit Phase (Detection)

Use a recursive Python script to scan the library for:

• Character Anchor Compliance: Verify presence of specific names (e.g., 'Sienna', 'Aura') and unique physical descriptors (e.g., 'Golden-Rust hair', 'hazel eyes'). Flag any use of generic terms ('a person', 'the girl', etc.).
• Technical Fidelity: Confirm mandatory gear strings are present (e.g., 'Arri Alexa 65', 'Zeiss Master Prime').
• File Integrity: Ensure every .json payload has a corresponding, synchronized .md file with matching filenames.

2. Rebuild Phase (Correction)

If failures > 0, execute a mass rebuild using an automation engine:

• Narrative Weaving: Do NOT simply append tags to the end of prompts. The script must rewrite/re-synthesize the prompt text so anchors are integrated into the descriptive narrative flow.
• Format Dual-Write: For every update, write both the .json payload (for API injection) and the .md file (for human review), ensuring headers match filenames exactly.

3. Verification Phase (Validation)

Run a final pass of the audit script to ensure:

• Compliance Rate = 100%.
• Total JSON Count == Total MD Count.

Pitfalls & Lessons Learned

• The "Structure vs. Content" Trap: A library can be structurally perfect (correct folders, filenames, and JSON syntax) while being semantically broken (missing character anchors). Always audit the content of the strings, not just the file metadata.
• Appendage vs. Integration: Appending tags like "Sienna, hazel eyes" to a prompt often results in lower-quality AI generations compared to weaving them into the prose (e.g., "Sienna, with her signature hazel eyes, gazes at...").

Reference Scripts

• audit_prompt_library_v7.py: The high-precision audit engine.
• rebuild_prompts_v2.py: The automated narrative weaving and sync engine.