name-search

NAME-SEARCH — marketing-grade product naming

The naming studio. Turns "I need a name" into a defensible, evidence-backed recommendation: a name — coined or real, any of the professional name-types — with no neighbours across npm/PyPI/crates/GitHub (and a clean domain), that reads right for the audience and archetype and encodes the philosophy. Cleaved out of ideate because naming is its own responsibility — ideate references this skill by capability when an idea needs a name.

The art-of-naming canon lives in ../../knowledge/naming/ (the taxonomy, phonosemantics, the affix-hook catalogue, the archetype/Neumeier/SMILE-SCRATCH frameworks, dev-tool conventions, ideation methods); this skill is the thin protocol that runs it.

The token contract (non-negotiable)

The LLM does only charter · generate · challenge · synthesize. Every availability/adoption check is the deterministic script — never spawn one agent per name (that pattern burned 20k+ tokens for a 50-candidate list). Call the script once for the whole candidate list, via at most one thin wrapper agent that returns its JSON:

bash ${CLAUDE_PLUGIN_ROOT}/skills/name-search/scripts/namecheck.sh \
  --json --adoption --neighbors --domains=com,dev,io,ai --connotation \
  --syllables=2-3 --registries=npm,github,pypi,crates  name1 name2 name3 ...

The script (scripts/namecheck.sh) checks npm (exact + hyphenated), PyPI, crates.io, and GitHub (user/org) in parallel, classifies each name CLEAR / LOW_ADOPTION / ABANDONED / TAKEN / UNKNOWN, and — opt-in — adds neighbour/typo-squat proximity (--neighbors: GitHub login-search, count==0 ⇒ no account and no neighbours), domain availability (--domains via RDAP), and a cross-language connotation screen (--connotation). Emits the JSON the report consumes, zero LLM tokens. Set GITHUB_TOKEN to lift rate limits (60→5000/hr core, 10→30/min search) — needed for a reliable --neighbors pass; without it the search API throttles and neighbour status is reported unknown (never upgraded).

How to run

The pipeline: discover ▸ charter ▸ generate (select veins) ▸ verify ▸ challenge ▸ synthesize.

1. Discover (interview). Follow references/discovery-protocol.md: infer everything you can from any upstream IDEA package / ideate output / bounded local trawl, then ask only the load-bearing gaps — audience, brand archetype, name-type appetite, power-adjacency, intellectual-humour appetite, risk appetite — one question per turn, each with a recommended answer + multiple-choice. Don't interrogate what's known.
2. Build the charter. Distil the interview into the enriched charter (references/charter-protocol.md): ranked values · audience & archetype · name-type strategy · risk/power-adjacency/humour · tone + exemplars · structural constraints · banned stems.
3. Select veins + generate a wide net. Follow references/generation.md: the charter's name-type strategy picks 4–8 veins from the full taxonomy (../../knowledge/naming/name-types.md) — suggestive, metaphor, mythological, compound, portmanteau, coined, acronym/backronym, scientific-taxonomic, animal, etc. — layered with the language/etymology veins, the affix-hook transform, phonosemantic tuning, and (opt-in) the power-adjacency + humour veins. 40–60 deduped candidates. Run one agent per vein, in parallel — never one per name.
4. Verify deterministically. Call the script ONCE for all candidates (the token contract above), with --neighbors --domains --connotation as the brief warrants. Filter to the recommendable set; record every name's full disposition (availability + neighbours + domains + connotation) for the report.
5. Challenge (scored). Follow references/evaluation-rubric.md: default survives=false; score each survivor on Neumeier 7, SMILE/SCRATCH, archetype-fit, sound-symbolism-fit, and cross-language connotation (cross-referencing the script's connotationFlags, neighbors, domains). Keep availability and challenge in separate verdicts.
6. Synthesize the report to references/report-template.md: brief summary, where-it-searched, attrition funnel, per-name disposition (incl. neighbour/domain columns) with the axis that killed each, per-vein contribution, the rubric scores, and a ranked shortlist with a top pick — confidence + residual risks stated honestly (trademark is web-search-only, RDAP is availability-not-legal; never asserted as cleared). Write to docs/marketing/naming-report.md (or stdout), and offer to render it via pressroom /publish.

Honesty rules

• availability ≠ challenge. A name removed for being taken (availability) is distinct from one demoted in the challenge. Keep the two verdicts in separate columns; never conflate "killed".
• UNKNOWN is not free. A registry whose probe could not complete is unknown, never upgraded to a verdict; say so and recommend a re-check.
• No silent caps. If the candidate list was truncated (--max-names) or a registry was skipped, say so in the report.

Self-improvement covenant

Carries the KAIZEN covenant (../../knowledge/covenant.md). When a name that passed the search later proves to collide or fall flat, that is not a one-off — it has a named home to be folded back into, via /ideator:self-improve: a missing name-type → a vein in ../../knowledge/naming/name-types.md; a missing hook → ../../knowledge/naming/affix-catalogue.md; a missing evaluation axis → references/evaluation-rubric.md; a missed collision/connotation → a check (or wordlist entry) in scripts/namecheck.sh / references/connotation-wordlist.tsv. Every future search, for all users, gets sharper by default.