Bots:
- Dolphin-Mistral generates uncensored/offensive prompts (first 100, then 5%)
- PrayBot_0 runs survival mode: auto-respawn, contextual low-health prayers
- Gemini 2.5 Flash Lite for diverse natural prompts
- Fixed Gemini markdown wrapper parsing, dolphin JSON format
POS printer:
- Triggers on $0.50 Gemini cost threshold instead of fixed interval
- Checks every 15 min, only prints when threshold crossed
- --check flag to see current cost status
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Prayer bots (ingame/prayer_bots.js):
- 3 Mineflayer bots that actively pray, sudo, and bug_log on dev server
- Gemini 2.5 Flash Lite generates diverse natural prompts on the fly
- Falls back to static pool if Gemini unavailable
- 15-45s interval per bot, 50/35/10/5 pray/sudo/bug/chat split
POS status printer (scripts/training_status_printer.py):
- Prints training data collection status to Epson TM-m30
- Tracks: dataset size, audit logs, bot activity, Gemini API cost, service status
- Triggers on $0.50 cost threshold (configurable), checks every 15 min
- --dry-run, --check, --force flags
Training:
- First LoRA run completed (233 examples, 3 epochs, loss 1.5→0.10)
- GGUF exported and loaded into Ollama as qwen3-8b-mc-lora on steel141
- Model is bad (expected) — hallucinating Chinese, leaking system prompt
- Deployed to dev server for live testing and data collection
- bf16 fix for Ampere GPU, system prompts included in training conversations
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- IDEA.md: project scope (Minecraft ops AI assistant via qwen3-coder LoRA/SFT)
- PLAN.md: complete roadmap with prior art analysis, architecture, phased plan, dev server docs
- data/schema.json: training example JSON Schema with negative_output support
- data/processed/seed_dataset.jsonl: 31 validated examples from repair code, prayer logs, session history
- data/validate_dataset.py: schema validator with summary statistics
- ingame/: Mineflayer bot framework (test_connect, spawn_bots, aware_bots with full event logging)
- Directory structure for knowledge/, eval/, training/, agent/ (Phase 1.3+ work)
Seth