feat: spark CLI v0.0.1 — initial implementation

[massimo92]

Summary

• Single bash script (spark) implementing all 11 commands from the spec
• Auto-profiler reads model config.json to detect reasoning parser, quantization, context length, MoE/multimodal architecture
• Interactive setup wizard with 3 phases (System Base, Remote Access, Inference Engine)
• install.sh curl-pipe-bash installer
• Full README documentation

Commands

setup · doctor · run · stop · pull · list · rm · status · logs · update · help

Key design decisions

• POSIX-compatible grep (no -P flag) for portability
• HF cache path resolution handles models--Org--Name double-dash convention
• GPU memory utilization auto-calculated from safetensors file sizes
• Profile caching at ~/.config/spark/profiles/ (bash-sourceable, editable)
• Config precedence: CLI flags > per-model profile > global config > defaults

Test plan

• spark help / spark version — basic CLI
• spark doctor — checks prerequisites
• spark list — lists HF cache models
• spark run --dry-run <model> — generates correct Docker command
• spark setup --check — read-only prerequisite check
• Full flow on GX10: pull → run → curl → stop

🤖 Generated with Claude Code