name-product

End-to-end product naming pipeline — brief → competitor scan → theme-by-theme generation → clearance → selection → TLD/budget → lock. Embeds the operator's default iterative shortlist-by-theme selecti

SKILL.md

GitHub

name-product

The Patently lesson + the ~50-name lesson. Naming "Patently" (an AI patent tool) collided with an 8-yr-old AI patent platform — a free domain proved nothing. And a diligent sweep of ~50 candidates showed real single words and common suffixes (-ly) are essentially gone in software/AI; the survivors were coined compounds and -proof forms. So: clear the name not the domain, and don't burn the whole budget hunting for a perfect real word that doesn't exist — but never assume a word is taken; check it.

When you activate

  • Naming a new product, or renaming one (before any domain purchase — hard gate in the go-live flow).

Core principles (carry these through every phase)

  • Positioning first. A name only feels right when pointed at a sharp positioning. Never generate in a vacuum.
  • Don't assume — check. Never write off a dictionary word as "taken" without checking. A taken .com is not a dead name; same-industry collision is what kills.
  • Keep the aperture wide. Explore real-related, real-unrelated (arbitrary), and made-up words. Bias toward arbitrary/coined for clearability and trademark strength, but don't pre-filter creativity.
  • Don't converge early. Diverge wide, then shortlist by direction; give the operator the wheel at each gate. (See [[feedback_dont_converge_early]].)
  • Keep two ledgers: a parking lot (liked-but-flagged — revisit, don't discard) and a rejected-names + why log (never re-pitch a killed name).

Pipeline

Phase 0 — Brief. Lock: core idea · ICP / target group · positioning (hero user, the one promise, the feeling, the enemy). Decide literal vs abstract. Save to products/<slug>/naming-brief.md.

Phase 1 — Competitor scan. Run the competitor-research skill. Outputs the naming conventions of the space, the white-space, and the collision baseline (names/roots to avoid).

Phase 2 — Theme select. Propose themes (territories) flowing from the positioning, each with sub-lanes: real-related · real-unrelated/arbitrary · made-up/coined · structural (suffixes like -ly/-ify/-o, "agent"/persona patterns, compounds, -proof, etc.). Exhaust one theme before the next, or have the operator pick the order. Confirm before generating.

Phase 3 — Generate wide. Within the chosen theme/lane, generate a generous batch (10–20). Creativity not pre-filtered. Include dictionary words — they get checked, not assumed.

Phase 4 — Clear. Run name-clearance: domain prefilter (signal, not gate) → same-industry collision → trademark signal (USPTO/EUIPO classes 9/42/45) → famous-crowding/SEO. Liked-but-flagged → parking lot. Killed → rejected log.

Phase 5 — Select (the operator's default method). Present cleared survivors grouped by theme in a compact table, then ask the operator to react/shortlist via multi-select — options, never a verdict. If a theme is exhausted with nothing loved, return to Phase 2 for the next theme. Iterate until there's a shortlist the operator actually likes.

Phase 6 — TLD + budget. For finalists, full TLD sweep (check-domains.ts) and align on budget (premium .com purchase vs accept .ai/.legal/.io). Confirm the operator's domain-tier preference.

Phase 7 — Lock. Recommend the strongest finalist, but the pick is the operator's. State plainly: a paid attorney trademark knockout (USPTO + jurisdiction, relevant classes) is required before committing. Then hand off to rebrand + domain attach.

Artifacts (persist per product)

  • products/<slug>/naming-brief.md — the Phase 0 brief.
  • products/<slug>/competitors.md — from competitor-research (compounds over time).
  • products/<slug>/naming-shortlist.md — parking lot + rejected-names-and-why + current shortlist.

Output

Always end a round with: the cleared shortlist (grouped by theme, with evidence + domain), the parking lot, and a single clear question giving the operator the wheel on what to explore next.