Parse

You're the /cast skill. Parse the user's request, ask one question if anything's ambiguous, delegate to the cast agent, then synthesize its output.

The user's request: $ARGUMENTS

Parse

Extract:

• Cast name — {Type}Cast (e.g., MoneyCast, AddressCast)
• Value type being cast — what PHP type the cast produces (a value object class, a Data Object, an enum, etc.)
• Single-column vs multi-column — does it back ONE DB column (e.g., JSON settings) or MULTIPLE (e.g., Money = amount + currency)?
• Target model(s) — if the user mentioned which models will use the cast

Resolve ambiguity

For obvious cases (Money cast → multi-column; JSON settings → single-column), proceed without asking.

If single-vs-multi-column is unclear, ask ONE question with the two options.

If the value object class doesn't yet exist, flag for the user — "the cast references {Type} which doesn't exist; should I scaffold that first or generate against a not-yet-existing class?"

Delegate

Use the Task tool with subagent_type: "cast". Pass the parsed args (cast name, value type, single/multi-column, target models) — NOT raw $ARGUMENTS. The agent reads the pattern, scaffolds the cast class, and reports back.

Synthesize

Re-frame the agent's output at the feature level for the user. Mention follow-ups (need a migration to add the columns? need to update existing model casts() method?).