skills.rest
vaskoyudha
vaskoyudhaCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

review-quality

Reviews competitive programming problems using adversarial Shield vs Sword analysis and scored quality criteria.

1|Updated Jul 17, 2026
Search Tags:#test coverage#scoring rubric#verdict#adversarial analysis#quality review#problem generation#competitive programming
One-click install
npx skills add https://github.com/vaskoyudha/agent-skills-problem-gen --skill review-quality-vaskoyudha
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: review-quality
Source: https://github.com/vaskoyudha/agent-skills-problem-gen/tree/main/.qwen/skills/review-quality
Command: npx skills add https://github.com/vaskoyudha/agent-skills-problem-gen --skill review-quality-vaskoyudha

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Competitive programming problems generated by AI pipelines often ship with hidden ambiguities, weak test coverage, or constraint mismatches that only surface during a live contest. This Skill acts as the final quality gate, catching these defects before publication through structured adversarial review. ## Core Features & Use Cases - Shield Scoring: Scores 10 quality dimensions (statement clarity, constraint alignment, solution correctness, test coverage, and more) from 0–10 with evidence, requiring all scores ≥ 8 for approval. - Sword Attack Vectors: Runs 5 adversarial attacks — ambiguity hunting, solution breaking, test bypass detection, constraint mismatch analysis, and edge case gap discovery — with severity classification. - Verdict & Revision Routing: Outputs a single APPROVED or REVISION verdict with a targeted revision owner (problem writer, solution engineer, or test generator) and actionable feedback, capped at 2 revision rounds. - Use Case: After a pipeline generates a problem statement, verified solution, and test suite, run this review to decide whether the package is contest-ready or needs a targeted fix round. ## Quick Start Review the generated problem package using the four upstream JSON outputs and return an APPROVED or REVISION verdict with scored criteria.

Frequently Asked Questions about review-quality

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a competitive programming problem for quality?▼

Score the problem across 10 criteria including statement clarity, constraint alignment, solution correctness, and test coverage, each from 0 to 10 with evidence. Then run adversarial attacks hunting for ambiguity, unsolvable cases, and test gaps before issuing an APPROVED or REVISION verdict.

What is Shield vs Sword adversarial review?▼

Shield vs Sword is a debate method where a defensive persona verifies quality standards with scored evidence while an offensive persona tries to break the problem. Sword wins ties, so any genuine doubt about a finding is treated as a real issue.

What inputs does the Gate 2 quality review require?▼

The review requires four upstream outputs: architect_spec.json, problem_draft.json, solution.json, and test_suite.json. If the solvability verdict is SOLVABILITY_FAILURE, the review halts immediately and returns REVISION targeting the problem writer.

When does a problem get APPROVED versus REVISION?▼

A problem is APPROVED only when all 10 Shield scores are at least 8, there are zero critical Sword findings, at most one defended major finding exists, and solvability passed. Any score below 8 or critical finding forces REVISION.

What happens if a problem fails review twice?▼

The review allows a maximum of 2 revision rounds. If round 2 still yields REVISION, the problem is force-approved with documented warnings rather than looping indefinitely.

Related Skills

Explore More Education & Research Skills→
Surya-SunkariSurya-Sunkari

solve

Generate and verify LaTeX mathematical proofs for Continuous Algorithms homework problems.

Community
Advanced
eroslifestyleeroslifestyle
1

python-patterns

Guide Python code with PEP 8, type hints, and design patterns.

Community
Intermediate
ok-helloworldok-helloworld
241

auth-agent

Detect authentication and authorization flaws in web and API targets.

Community
Advanced
sota411sota411
1

adversarial-panel

Runs isolated multi-model Codex panels with blind answers, cross-critique, and evidence-weighted synthesis.

Community
Advanced