make-montage

Build a beat-synced montage from the arguments: $0 is the video path, $1 is the audio path, $2 is the optional output path (default: montage.mp4 in the current working directory).

How this skill thinks

Cadence runs as a three-party conversation: you (Claude) orchestrate tool calls; Gemini acts as the subject-matter expert for video/audio perception and creative judgment; the user speaks through you. Gemini's reasoning carries across the calls in a session — it remembers what moments it found, which segment it picked, and why. Your job is to synthesize the user's raw request into a coherent userIntent at session-begin, then make targeted calls and inspect results.

Interpreting the user's request

Split intent across two axes before calling tools:

• Content intent — what should appear in the montage. Feeds the session's userIntent (shapes every Gemini call) and the optional per-call focusPrompt for refinement.
• Arrangement style — how the content should be cut. Passed as stylePrompt to reason-plan-edit.

Enrich vague user phrases into concrete criteria before passing. "Badass close combat" → "physical melee engagements — punches, kicks, grapples, throws, blocks; exclude ranged attacks, environmental destruction, and dialogue."

If the user hasn't given arrangement direction, ask. Reasonable defaults: