What problem does it solve? AI-generated music often sounds generic because no deliberate compositional choices were made. This Skill replaces prompt guessing with a structured ARR-SPEC arrangement specification—key, tempo, bar-accurate section map, harmony, instrumentation entry/exit, energy curve, vocal delivery, and mix intent—that is compiled into Suno prompts, ABC scores for YuE2, or MIDI, and then used as the acceptance baseline for the generated audio. ## Core Features & Use Cases - S0–S7 Stage Workflow: Gates each phase from intent definition through skeleton, material development, harmony, arrangement, vocal/mix intent, compilation, generation, and acceptance, with explicit pass criteria per stage. - Skill Routing & Boundary Table: Maps every compositional decision to exactly one owning skill (harmony, form, orchestration, vocal direction, style layers, etc.) so no two skills claim the same decision. - 20-Point Self-Audit (SPEC-LINT): A manual checklist that counters AI-generation defaults such as uniform 8-bar sections, monotonically rising energy, grid-locked timing, and identical chorus repeats. - Use Case: Ask for a city pop track; the agent fills an ARR-SPEC with asymmetric bar counts, a descending energy point before the final chorus, per-instrument push/pull timing offsets, and a style prompt, then compiles it for Suno or YuE2 and checks the result field-by-field against the spec. ## Quick Start Ask the agent to compose a new song from a brief and have it load this workflow first to produce a complete ARR-SPEC before any generation.
