quest-solution-review
GitHub用于对比社区挑战(Quest)中多个竞争Pull Request,依据正确性和代码简洁性筛选出最优实现方案。通过验证验收标准、外部契约及安全性,评估候选方案的复杂度与测试覆盖度,最终输出包含排除理由、阻塞项和合并建议的评审报告。
Trigger Scenarios
Install
npx skills add pollinations/pollinations --skill quest-solution-review -g -y
SKILL.md
Frontmatter
{
"name": "quest-solution-review",
"description": "Compare competing pull requests for a Pollinations community quest and select the smallest correct solution. Use when reviewing multiple implementations linked to a POLLEN-QUEST issue."
}
Quest Solution Review
Choose the implementation that fully solves the quest with the least ongoing
complexity. Do not reward being first, large, or already up to date with main.
Review
- Read the quest, acceptance criteria, and relevant current code.
- Find every linked implementation from the issue timeline, closing keywords, cross-references, and PR search. Include closed and draft PRs when they are genuine submissions.
- Inspect each complete diff and checks. Separate branch-age conflicts from the quality of the proposed change.
- Verify changing external contracts against current primary documentation.
- Apply the gates below, then test the strongest candidates in proportion to risk. Prefer focused tests of real code and a practical smoke test when one is feasible.
Correctness gate
A candidate must:
- Meet the quest's observable acceptance criteria.
- Use the real provider/API contract and the repository's existing architecture.
- Preserve required auth, permissions, security, accounting, and tracking.
- Bill from the provider's reported usage when the route reports it, not from a price formula rebuilt from request parameters.
- Handle failure modes that users can realistically encounter.
Reject a candidate that only appears to work, silently changes the public contract, or requires speculative fallback behavior to cover a defect.
Simplicity comparison
Among correct candidates, prefer:
- The smallest focused diff and fewest new concepts.
- Existing shared utilities and established data flow.
- Clear ownership and direct code over wrappers or generic frameworks for one use case.
- Focused tests that exercise the production implementation.
- Registry, schema, and documentation changes only when the public contract requires them.
Treat unrelated edits, generated noise, lockfile churn, duplicated helpers, hard-coded copies of dynamic data, mock infrastructure, and contaminated commit history as costs. Line count is evidence, not the decision by itself.
Report
Give a concise comparison containing:
- Candidate PRs reviewed and any submissions excluded, with the reason.
- Blocking correctness findings.
- Relative complexity and useful ideas found in alternatives.
- Tests, checks, and smoke tests actually run.
- A verdict: merge, request a small follow-up, combine a specific idea, or accept none.
Do not comment, close, edit, or merge contributor PRs unless the user asks. If a maintainer follow-up is authorized, keep it minimal and preserve contributor attribution. A small bonus for an alternative is exceptional: use it only when the alternative was independently strong or materially influenced the shipped solution, not merely because it was submitted.
Version History
-
7d1d825
Current 2026-09-28 05:13
修正了z-image fal回退计费逻辑,改为从fal响应获取实际计费的兆像素数,避免估算导致的成本低估;同时更新了AGENTS.md和技能文档中的计费规则说明。
- 3e1cabf 2026-09-09 11:59


