brainstorm-tournament

/brainstorm-tournament — diverse generation + rubric-scored pairwise tournament

Runs the bundled grok-workflows harness, which generates candidates from several distinct angles (literal, evocative, playful, technical) in fresh agent contexts so the pool is genuinely diverse, dedupes them, derives or accepts a judging rubric, and runs a pairwise tournament (one agent per match) to find #1, #2, and #3. Pairwise comparison beats noisy absolute scoring for taste calls. You do not re-implement any of this — you invoke the harness and act on its JSON.

Usage

/brainstorm-tournament <thing to name/design> [:: rubric]

Everything before :: is the subject; anything after :: is an optional user-supplied rubric. If no rubric is given, the harness derives one.

How it runs

This skill bundles an entrypoint at <skill-dir>/scripts/run.mjs (thin delegator to the centralized launcher logic in src/launcher.mjs; self-location still works via the passed import.meta.url from the delegator) — <skill-dir> is this skill's own directory, whose absolute path is announced in your system context when the skill loads. Derive the entrypoint path from that announced SKILL.md path and inline the absolute path into a single run_terminal_cmd call (don't rely on the working directory or a shell variable). The launcher locates its bundled harness itself (via the delegator), so no repository path is needed:

node <skill-dir>/scripts/run.mjs "<thing to name/design> :: <optional rubric>"

The harness prints a single JSON object to stdout (progress logs go to stderr):

{
  "rubric": "the rubric that was applied (user-supplied or derived)",
  "top3": [
    { "candidate": "Winner name", "why": "one-line deciding reason" },
    { "candidate": "Runner-up", "why": "..." },
    { "candidate": "Third place", "why": "..." }
  ],
  "poolSize": 24
}

What to do with the result

1. Parse the JSON from stdout.
2. Present top3 as the primary answer — list each candidate with its why, in rank order (#1 first). Lead with the #1 pick.
3. State which rubric was applied so the user can see the basis for the ranking, and mention poolSize (how many unique candidates were considered) so they know the shortlist came from a broad pool, not a silent cap.
4. If top3 is empty (poolSize 0), tell the user no candidates survived and suggest rephrasing the subject or supplying a clearer rubric, then re-run.

Do not invent or re-rank candidates yourself — the tournament ranking is the point of this skill. If the user wants more than three, re-run with a refined subject or rubric rather than padding the list.