Name Cracker

Run a broad, structured name search for the thing the user wants to name. This skill is the orchestrator role: it interviews the user into a naming brief, writes one shared generator brief file to the run's scratch directory, spawns one generator subagent per naming-strategy file under references/categories/ — each pointed at the shared brief instead of receiving it verbatim — verifies every generator's candidate file on disk, merges them with scripts/merge-candidates.mjs, curates a shortlist, and delivers one file the user chooses from. The orchestrator never generates names inline — diversity comes from independent contexts, each anchored to a single strategy.

Subagent capability precondition

This skill REQUIRES a runtime primitive that spawns independent subagents with their own context windows and lets them write files to disk before replying. There is no inline fallback: one context generating every category back-to-back converges on samey names and defeats the fan-out. If the runtime does not support subagents, stop and tell the user this skill cannot run here.

Inputs

All optional:

• Output path — where the deliverable is written. Default: ./name-candidates.md.
• Candidates per category — the maximum each generator may propose. Default: 3.
• Category subset — restrict the run to named categories or themes (e.g. "just mythology, metaphors, and puns"), matched against the filenames and H1 titles under references/categories/. Default: every category file.

Procedure

1. Interview the user. Generated names are only as good as the understanding of what is being named, so the interview comes before everything else. Open by asking what the thing is and does, then follow up — a few questions at a time, adapting to the answers — until every line below is filled or explicitly empty; skip only lines the invocation already answers: