Compare competing pull requests for a Pollinations community quest and select the smallest correct solution.
SkillScore100Maintained
Automatic
◆Coding pack
Last verified 10/10/2026
What it changes
Passed 19 of 19 test cases with the skill, and 18 of 19 without it.
Without skill95%
With skill100%
Checks it passed that the model alone missed
Avoids using line count as the primary decision factor.
1 fixed18 passed either way0 still failing0 broke3 errored
Pass-rate lift
+5 pp
Tokens
+7%
Agent turns
0%
Response time
-27%
Model gemini-3.6-flashMeasured 10/10/2026v319 of 22 cases passed with skill · 19 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
A test case the skill fixed. Same model, same prompt, with and without the skill.
Prompt
PR #111 for quest #60 has 45 commits and 800 lines of trial-and-error code. PR #112 has 2 commits and 40 lines of clean code that addresses all review feedback after thorough local testing. Compare these candidates in a review report.
What was checked, without and with the skill
Checked
Without skill
With skill
Avoids using line count as the primary decision factor.
Missed
Passed
Recommends PR #112 for superior maintainability.
Passed
Passed
case-16 · gemini-3.6-flash · 10/10/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md69 linessingle file
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.
Create and post ASCII art voting status diagrams to GitHub issues and Discord. Use when asked to update voting, show voting results, or announce voting status.