brainstorm-and-refine
Use when asked to brainstorm then refine, generate ideas then improve, or explore alternatives and synthesize the best one across passes.
- License 许可
- MIT
- 所属仓库
- tony/skills
- 最近更新
- 2026-11-08
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- brainstorm-and-refine
- description
- Use when asked to brainstorm then refine, generate ideas then improve, or explore alternatives and synthesize the best one across passes.
- allowed-tools
- [ "Bash", "Read", "Grep", "Glob", "Write", "Task", "AskUserQuestion" ]
- argument-hint
- [--variants=N] [--passes=N] [--timeout=N|none] [--mode=fast|balanced|deep] [--judge=host|round-robin] [--preamble=...] [--no-deslop|--quiet-deslop|--verbose-deslop] [--workers=subagents|model-clis] <prompt>
02 资源文件 1
- 参考文档agents/openai.yaml268 B
03 SKILL.md 正文
Weave Brainstorm and Refine
Generate independent originals from adversarial participants, let the user choose which enter refinement, then refine them: a judge picks the best, the weave folds in the runners-up's strengths, and the woven result goes back to every participant for another round.
The prompt comes from $ARGUMENTS. If there are none, ask the user what to brainstorm and refine.
<engine> is ../../scripts/engine.py resolved from this skill's directory to an absolute path, written literally in every command, and run with python3.
Worker selection
<!-- portable: ask-user-choice=headless-default -->
Before any other unresolved configuration choice or operational step, read
../../references/worker-backends.md (Caller contract). Resolve
worker_backend from --workers=subagents|model-clis using that reference;
if the flag is absent, ask its worker question first.
If interactive choice is unavailable, honor its documented headless default.
Phase 1: Start the session
Follow ../../references/ensemble-core.md (Start the session) with --command brainstorm-and-refine, resolving variants per ../../references/variants.md. The pass count covers the pass that judges the originals, so passes 1 judges and weaves once.
Phase 2: Brainstorm the originals
Dispatch the unchanged originals prompt of ../../references/variants.md per ensemble-core.md (Dispatch lanes), then (Close the repo guard).
Phase 3: Show the originals and ask which to refine
Show every original as ../../references/variants.md step 5 says.
Ask through AskUserQuestion which originals enter refinement:
- All of them (Recommended): every successful original.
- Let me pick: a follow-up multi-select with one option per original, such as
claude-v1. - None: stop after the brainstorm.
Without AskUserQuestion, refine all. Never start refinement without this step. Record the choice:
$ python3 <engine> update \
--session "<session_dir>" \
--event brainstorm_to_refine \
--event-json '{"selected": ["<id>"], "available": <N>}'
On None, follow ensemble-core.md (Finish) and stop with the brainstorm outputs.
Phase 4: Refine
Follow ../../references/refine-cycle.md for pass 1, with the selected originals in the session's outputs/ directory as the candidates. Judge pass 1 with judge mode host even under round-robin, because originals vary in structure; later passes use judge. The weave goes to pass-0001/woven.md.
For each later pass up to passes, the cycle's Distribute step has already written the prompt. Dispatch it per ensemble-core.md (Dispatch lanes) with the new pass number, close the repo guard, then run the cycle again with that pass's outputs/ as the candidates.
Phase 5: Deslop and present
Follow ../../references/refine-cycle.md (Deslop, present, and finish) with ../../references/deslop-pass.md, ../../references/present-results.md, RESULT_KIND = brainstorm-and-refine and ../../references/results/brainstorm-and-refine.md. The reply starts with the hero line alone on its own line, with no clause appended, and has only the four headings of that kind file: the woven design goes under Final Result as paragraphs and flat bullets, with its own headings and bold labels folded into the first words of a bullet and no label line of its own. The Evolution Summary's Originals bullet comes from the Phase 3 originals: how many there were and what they differed on. Write the reply per ../../references/prose-rules.md.
Rules
04 同仓库更多 Skills
serene-bliss
Use when asked to make a CLI --help, docs page, error message, or tooling UX more serene or calmer for developer experience; returns text, no edits.
review
Weave code review: run independent adversarial reviews in parallel, then synthesize findings
refine
Use when asked to refine a draft, iterate on code until cleaner, or polish a README through repeated critique and improvement.
prompt
Weave prompt: compare independent adversarial implementations, then adopt the single best one
plan
Weave planning: compare independent adversarial plans, then synthesize the best one
fix-review
Fix weave review findings: validate, add test coverage, fix, and commit each as atomic changes