flora-mcp-skills: Skill for Claude Code

flora-batch-generate

What it is

Not a faster chatbot. A pipeline. The user hands over a list; the list comes back as assets. Nothing in between is a conversation.

The difference is not politeness, it is arithmetic. Forty items generated one at a time, each one waited on before the next is asked for, is forty sequential waits. The same forty fired together is one wait. FLORA runs them concurrently either way — the serialisation is entirely on the agent's side, and it is the single thing that decides whether a batch takes ninety seconds or an hour.

Measured, this skill's own hand-test: six items, fired together, all complete in 15.8 seconds. The same six at the model's estimated 25s each, run one after another, would be about 150 seconds. Same credits, same outputs, one tenth the wall clock.

The law

N items is one wait, not N waits. Fire every item before you poll any item. Poll the batch, never the item.

Both halves matter and they fail differently. Firing serially wastes the user's time in the open. Polling serially wastes it invisibly — the batch is already running fine and the agent is the bottleneck.

Everything below is in service of that one sentence.

Inputs

ITEMS     the list — rows, SKUs, variants, URLs, filenames   required
STYLE     the one treatment every item shares                required
MODEL     one model for the whole batch                      optional, see Planning
PROJECT   where the outputs land                             optional, one is created

One style, N items. This is the shape the skill exists for: a single consistent treatment, varied only by the per-item variable. A product list becomes PDP shots by holding the lighting, backdrop, lens and framing fixed and changing only the subject noun. If two items genuinely need different treatments, that is two batches — say so rather than quietly blending them, because a blended batch has no consistent style and the user cannot tell which axis moved.