st-gen-image-promptlisted

# ST Gen Image Prompt — Magnum-free image prompt builder Generate complete booru-tag image prompt cho SillyTavern. Replaces Magnum Mode 4 extraction. Output paste-ready for ST `🎨 Freestyle` button (Mode FREE=6 pass-through). **Workflow:** 1. Read chat context (last N msgs từ ST chat .jsonl) HOẶC `--describe '<custom>'` 2. Read identity baseline từ `data/identity-baselines/<CharName>.txt` 3. Generate booru tags theo NoobAI XL conventions 4. Verify tags qua Danbooru DB (lazy fetch ~5MB CSV vào `~/.cache/`) 5. Output paste-ready prompt + verification report ## Constants ``` ST_DATA = /home/haint/Projects/home-server/sillytavern/data/default-user SKILL_DIR = ~/.claude/skills/st-gen-image-prompt SKILL_DATA = $SKILL_DIR/data CACHE_DIR = ~/.cache/st-gen-image-prompt DEFAULT_LAST_N = 10 TAG_DB_URL = https://raw.githubusercontent.com/DominikDoom/a1111-sd-webui-tagcomplete/main/tags/danbooru.csv TAG_DB_TTL_DAYS = 30 MIN_TAG_COUNT = 50 ``` --- ## Phase 0: Parse Arguments Extract from `$ARGUMENTS`: - `CharName` = first non-flag positional. If absent → auto-detect from latest `chats/<dir>/*.jsonl` mtime. - `--describe '<text>'` → custom scene (overrides chat reading) - `--last N` → number of chat messages to read (default 10) - `--no-clipboard` → skip auto-copy to clipboard (default = auto-copy enabled) Validate: - Char folder `chats/<CharName>/` exists OR `--describe` given. If neither → error. --- ## Phase 1: Gather Context **Step 0 — Ensure tag DB cache:** ```python import